Why You Should NOT Use ARIMA to Forecast Demand

Are you using (S)ARIMA(X) to forecast supply chain demand?
I see five reasons why you should not.

1. 💾 ARIMA requires a long historical horizon, especially for seasonal products. Using three years of historical demand is likely not to be enough. Short Life-Cycle Products. Products with a short life-cycle won’t benefit from this much data. Forecasting demand at a higher hierarchical level might help. But it will come with other challenges (reconciliation, loss of accuracy).
2. 💻 Running ARIMA on a wide dataset is (extremely) time-consuming as each SKU needs to be optimized separately. If it takes 1 second to optimize one SKU, that’s nearly 3 hours for 10,000 SKUs. You will need a lot of computing power (and parallelization).
3. 📐 ARIMA is assuming time series to be “Stationary.” It means that ARIMA is assuming your demand to have a constant mean, variance, and covariance over time. Let me know if this description matches any of your products.
4. 📈 At its core, ARIMA is powered by linear regressions. ARIMA assumes each product's trend to be constant over time. Would you agree that this is the case for all your products?
   Other statistical models like exponential smoothing with (damped) trends can deal with changing trends. That’s much more realistic.