Hi, potentially a more theoretical question but here goes. I am forecasting for retail stock replenishment/demand planning. I have a broad assortment with various lead times (so evaluation at SKU level may be over different horizon lengths). However, another consideration is not only the lead time forecast but also the (e.g. end of Season) closing stock position, which is often rolled up to Category (higher agg level). One approach I considered was a hierarchical solution, for example create a long term Category-level forecast (evaluated over the remainder of the Season/longer horizon) that is reconciled with SKU forecasts (evaluated over the lead time). I am not sure if this is possible in HierarchicalForecast, or even a sensible approach? I suspect as a bare minimum, I would need to create a forecast over the long term at SKU level so that it can reconcile effectively with the Category level forecast? I am not against providing SKU forecasts over the longer horizon (i.e. in line with the Category level horizon) but I would like to ensure that they've been evaluated in line with their primary use case. Thanks!

I'd create forecasts at the same time granularity (unfortunately temporal reconciliation is not yet available, we're working on that) for the same time steps and perform a hierarchical reconciliation. I.e., I don't see an issue with the approach

Thanks @Olivier for the response. Apologies if I was unclear, I am considering the same temporal granularity (weekly data) at both levels, but the horizon of interest is different. For example, at SKU level the forecast horizon of interest might be 4 weeks (i.e. lead time for the SKU replenishment decision) but the Category level forecast is until the end the current Season, say 20 weeks away. When I select the winning forecast (via backtesting/cross-validation) I want to ensure that the model selected is the most accurate over a 4 week horizon (primary use case) but has something sensible/decent over the longer term (secondary use case). To do this, one approach might be to hierarchically reconcile the best Category forecast optimised/selected over h=20, and a SKU level forecast optimised over h=4 (but a 20 week SKU forecast is provided to support reconciliation). Would this be a sensible approach?

Yes! I don't know what models you're using but if you're using MLForecast you can use a separate model for each horizon. In that case, there should be almost no performance difference between having h=4 and h=20 for the SKU model. But I'd just start with a model that can do h=20 but is optimized for h=4, for example by applying a horizon_weight (if using NeuralForecast, for example).

Thank you so much @Olivier!