1. Abstract
Reliable analysis in Board Foresight starts with datasets that meet minimum standards for history, completeness, and consistency. Ensuring that data satisfies these requirements before it is explored or modeled helps produce stable insights, clearer correlations, and more trustworthy forecasts.
2. Context
Apply this best practice before using any dataset in Discover, adding it to a Workbench, or including it in a Predict job. These checks should be part of standard data preparation for all time-series indicators, whether internal or external.
3. Content
3.1 Why It Matters
Time-series analysis relies on patterns observed over time. Datasets with limited history, excessive gaps, or inconsistent structure often produce misleading insights, even if tools technically allow them to be used.
When minimum data requirements are not met:
- Correlations may appear strong but fail to hold over time
- Trends may reflect short-term noise rather than underlying behavior
- Models may overfit recent periods and perform poorly out of sample
Applying minimum data standards ensures that insights drawn from Discover, Workbenches, and Predict are grounded in sufficient historical context and are more likely to generalize as conditions change.
3.2 How to Apply
Before using a dataset in analysis, confirm that it meets the following minimum requirements:
- History
- At least 5 years of data, or 60 monthly observations
- This provides exposure to multiple business and economic cycles
- 3 years of monthly data is the minimum viable history for use in the Discover and Predict engines.
- Completeness
- Fewer than 5% missing values across the full series
- Any missing periods should be addressed prior to data load using documented, business-aware methods
- Frequency
- Monthly data is preferred for most analyses
- Quarterly data may be acceptable but requires longer history to achieve the necessary number of observations
- Continuity
- No gaps in the time series
- Quality
- Values fall within reasonable business ranges
- Units, scales, and definitions are consistent over time
If a dataset does not meet these criteria, consider extending it with historical archives, supplementing it with related indicators, or excluding it from analysis until requirements are met.
3.3 Example
An analyst wants to include a retail sales indicator that contains only 36 months of history. Before modeling, they locate archived financial reports and extend the series to cover five full years. With sufficient history now available, the indicator can be confidently used in Predict.
3.4 Common Pitfalls
- Accepting short histories because the model “still runs”
- Mixing monthly and quarterly indicators in the same Workbench without adjustment
- Ignoring small but recurring gaps that truncate training windows
- Using datasets with silent definition or unit changes over time
3.5 Expected Results
- More stable coefficients and cleaner residuals
- Forecasts that generalize better to future periods
- Reduced need for rework or model rebuilds
- Increased confidence in model outputs and decisions based on them
