Forecasting perishable commodity prices is inherently difficult due to strong volatility and intricate seasonal patterns. Therefore, forecasting models should effectively capture the underlying structure of the data. This study focuses on forecasting daily onion prices across three key Indian cities using open-source data from 2020 to 2024. The dataset includes daily observations of minimum, maximum, modal, and retail prices, along with market arrival quantities. The modal price, representing the most frequent transaction value, was selected as the primary forecasting variable. Three univariate time series models were applied to assess their forecasting performance. Univariate models were used to isolate forecasting performance and evaluate the effectiveness of decomposition in stabilizing highly volatile daily price series. These models include Seasonal Autoregressive Integrated Moving Average (SARIMA), Seasonal and Trend decomposition using Loess combined with ARIMA (STL-ARIMA), and STL combined with Exponential Smoothing State Space models (STL-ETS). Model performance was evaluated using standard accuracy metrics such as RMSE, MAE and MAPE. Results show that the STL-ETS model consistently achieved the lowest forecast errors across all three cities, outperforming both SARIMA and STL-ARIMA. The findings highlight the suitability of decomposition-based models for capturing high-frequency variations in agricultural price data. This research establishes an evidence-based comparison of univariate forecasting models and demonstrates that decomposition-based methods offer superior accuracy for highly volatile perishable price series.