In the original notebook, each monthly observation is dated at the start of the reference month (e.g., data dated March 1st represents March's activity). In practice, these figures are only published 1–8 weeks after month-end.
To reflect real-time availability, the original analysis manually shifted each series forward based on its typical publication lag (see Table 1 in the paper):
| Release lag (from month-end) | Shift applied | |---|---| | < 30 days | +1 month | | 30–60 days | +2 months | | > 60 days | +3 months |
This adjustment currently exists only in the original notebook and is not implemented in the Shiny app or the refactored codebase. As a result, users who input raw data bypass this correction, which may produce optimistic nowcast accuracy (look-ahead bias).
Add a preprocessing step that: