currently I'm having a look at the wonderful hts-Package. Its idea is to forecast hierarchical time series (like described here). In this example the unit of measurement across all time series is the same, e.g. people.
Does this approach (especially the
combinef-Function in optimal reconciliation mode) also works for mixed unit time series? I'm thinking for example of weather-data where the temperature of one time series is measured in C° and of another in F° (same quantity, different unit). Or even more complex: Currency in different countries with daily changing exchange rates.
Aggregating such data would at least render the level 0 result useless, but does it work on the level 1, assuming that all levels below 1 use the same unit (e.g. $ or F°)?