1. Abstract
Large one-off shocks, such as strikes, pandemics, or natural disasters, can distort time-series data. Adding control binaries isolates these events, ensuring your model focuses on true economic relationships.
2. Context
Apply when a dataset contains periods of extraordinary disruption unrelated to normal market activity, before model estimation or scenario testing.
3. Content
3.1 Why It Matters
Ignoring one-off shocks leads to biased coefficients and misleading forecasts. Explicitly controlling these anomalies preserves model integrity.
3.2 How to Apply
- Identify disrupted periods (for example, March–May 2020 COVID lockdowns) by looking for significant outliers in the residuals chart in the Diagnostics tab of the model.
- Search for existing control binaries (e.g., “Pantry Loading / Initial COVID Shock.”) or upload your own custom binary using the Data Import Wizard.
- Confirm its values: 1 during the event, 0 otherwise.
- Add the binary to your model.
- Review Diagnostics to verify reduced residual distortion.
- Record the change in documentation.
3.3 Example
A manufacturer adds a binary for a 2021 factory shutdown. After inclusion, the model’s residuals normalize and no longer over-predict losses.
3.4 Common Pitfalls
- Treating recurring seasonal swings as one-time shocks.
- Misaligning binary timing with data frequency.
- Removing control variables after successful validation.
- Using too many binaries.
3.5 Expected Results
Models that remain stable and explainable even when the data includes major non-economic disruptions.
