Forecasting and model validation
Forecasting methods range from driver-based accounting logic to statistical and neural approaches; the choice depends on the data, the horizon, and whether meaningful drivers exist.
- Operational use
- Balance-sheet and cash forecasting, demand and capacity planning, scenario comparison for management decisions.
- When not to use it
- When the business has direct knowledge a model cannot see (a signed contract, a planned action), deterministic driver logic beats statistical extrapolation. Sophisticated models must first beat naive baselines to earn their complexity.
- How it is validated
- Backtesting against held-out history; comparison to simple baselines; error measures appropriate to the decision; assumptions and intervals shown, not hidden.
- Limitations
- A forecast is a decision tool with an error range, not an accounting fact. Neural methods such as N-BEATS and DeepAR can excel on rich series and fail on short, structural, driver-dominated business data.
The forecasting literature is rich: N-BEATS demonstrated that carefully designed neural architectures can win on pure time-series benchmarks; DeepAR made probabilistic forecasts practical at retail scale. Neither result means a neural model is the right tool for a specific company’s cash forecast.
Eigenvalue’s method selection starts from the data and the decision: how much history exists, whether known drivers dominate, what horizon matters, and what error is tolerable. Statistical estimation earns its place through backtests against baselines. And in financial forecasting specifically, the structural layer (accounts that must balance, entities that must tie) is not optional modeling flavor; it is the difference between a forecast and a number.
Where it appears in our work
- Quantitative Forecasting and Scenario Engine Implemented in this system
Primary sources
- Oreshkin et al. (2019). N-BEATS: Neural Basis Expansion Analysis for Interpretable Time Series Forecasting
- Salinas, Flunkert, and Gasthaus (2017). DeepAR: Probabilistic Forecasting with Autoregressive Recurrent Networks
Last reviewed 2026-08-02. Sources link to primary or official pages; verify current versions before relying on them.