Index-based stock assessments often use time series smoothing to reduce estimation error in their output. However, this may increase error if actual changes in the population are masked by the smoothing. Therefore, it is important to understand how smoothing assessment output will affect estimation accuracy under conditions commonly observed in fish survey time series. Here, we simulated the Georges Bank yellowtail flounder survey time series and assessment in which three bottom trawl surveys are used to estimate the state of the population. We compared the accuracy of the assessment output when it is unsmoothed versus when is smoothed by applying a three-year moving average to the output index. We did so for two estimation models: the currently used empirical assessment model, and a multivariate state-space random walk model. We simulated five biomass trends under two drivers, and examined model performance when one, two, or three surveys were used to fit the model. Overall, the unsmoothed state-space model consistently outperformed the other methods, particularly when there was a rapid change in biomass in the final years, and all three surveys were used to fit the model. The benefit of using the unsmoothed estimate also increased with the underlying rate of change in the population. The advantage gained by using the unsmoothed estimate also increased as additional surveys were used to fit the model. The benefit of using the state-space model versus the empirical model was similar to the benefit of using the unsmoothed versus smoothed estimate. Our results suggest that an unsmoothed state-space model should be used when multiple survey time series are available and population biomass appears to be rapidly changing in recent years.