Estimating and forecasting the term structure of interest rate is a challenging task since various models(Vasicek, Clk, Nelson Siegel, etc.) exist and all models have their merits and demerits. For example, the Vasicek and Clk models, as opposed to the Nelson Siegel model, impose the desirable absence of arbitrage restriction; however, the Nelson Siegel model renders a better empirical fit as compared to the Clk and Vasicek models. The Arbitrage Free Nelson Siegelmodel by combining the empirical attractiveness of the original Nelson Siegel model and the no-arbitrage condition is considered better in terms of in-sample fit and forecasting performance to the original Nelson Siegel model. The study aims to compare the in-sample fit and out-sample forecasting performance of the three-factor Vasicek, Clk, and AFNS models in the state-space framework in the context of Pakistan's bond market, a scenario rarely considered in the literature. The study concludes that the 3 factor Clk model has better in-sample performance than the AFNS and Vasicek models. Moreover, the Random Walk model has better forecasting performance than all three models, but the Clk model has better out-sample results than the AFNS and Vasicek models. The Vasicek model has the poorest performance compared to all models in terms of in-sample and out-sample