Abstract
Reliable forecasts of carbon price fluctuations can help firms plan allowance purchases as emissions trading expands. We examine whether allowance trading volume, the quantity of emission permits traded, improves forecasts of the size of the next recorded price change, regardless of its direction. We compare models using price information with otherwise identical models that also use trading volume. Forecasts cover 17 allowance instruments in nine Chinese markets. The methods are ridge regression, a model based on generalised autoregressive conditional heteroskedasticity (GARCH), random forests, and their average. During 2019–2023, adding trading volume reduces the mean absolute error (MAE) by 4.33% for the GARCH model and 1.39% for the random forest (both adjusted p < 0.001). In 2024–2025, it increases MAE in all four models, with no reliable benefits in the sensitivity checks. Models trained on regional markets also show no reliable benefit from trading volume when forecasting national allowance price fluctuations. However, when underestimating a movement receives five times the penalty of overestimating it, trading volume improves three models. Its usefulness therefore depends on the market, period, and consequences of forecast errors. Firms should test these conditions before including trading volume in risk forecasts.