Yes, Transformers are Effective for Time Series Forecasting (+ Autoformer)

yeah, transformers can work for time series, but don't get starry-eyed about them being a magic bullet. they're effective, sure, but they introduce a lot of overhead, especially when you're dealing with typical industrial forecasting.

autoformer helps with the complexity of the long-range dependencies, which is useful, but the computational cost and the sheer amount of data required just to set up the architecture makes it less practical than simpler ARIMA models for many use cases.

we need to weigh the theoretical performance gains against the real-world inference latency and training time. if you need fast, deployable models, simpler approaches might still win over the full transformer stack.