<![CDATA[This study aims to determine whether daily television audience dynamics exhibit statistically significant conditional variance dependence that is systematically overlooked in conventional ARIMA-based broadcasting forecasts and to assess the incremental empirical value of integrating ARIMA with GARCH modeling. Using 1,096 consecutive daily observations (2022–2024) of viewers for a nationally broadcast program, we implement a diagnostic-first framework that jointly evaluates conditional mean and variance processes. Stationarity is confirmed through the Augmented Dickey–Fuller test (ADF = −3.4693, p = 0.0088), and an MA(1) specification is selected for the conditional mean (AIC = 1302.76). Residual diagnostics reveal pronounced ARCH effects (ARCH-LM = 78.4602, p < 0.001), justifying second-moment modeling. Among competing variance specifications, GARCH(2,2) yields the lowest information criterion (AIC = 1060.321) and indicates near-unit volatility persistence (Σα + Σβ = 0.9856), evidencing durable intertemporal uncertainty transmission. Out-of-sample forecast evaluation demonstrates low relative error (MAPE ≈ 1.0%), supporting empirical robustness. Unlike prior ARIMA-centered broadcasting studies that prioritize point accuracy under homoscedastic assumptions, this integration explicitly models volatility clustering as an object of inference, aligning media analytics with established volatility frameworks without overstating cross-domain novelty. The findings show that incorporating conditional variance dynamics provides measurable gains in risk-sensitive forecasting, offering a replicable approach for advertising allocation and scheduling decisions in competitive media environments.]]>
