InteractiveMap of Operations Research
Back to Foundations of OR
MethodFoundations01.09
Foundations of OR

Forecasting for OR

Turn historical data into decision inputs, not just predictions.

Overview

Forecasting for OR turns historical data, domain judgment, and uncertainty estimates into inputs for prescriptive models. The forecast is not the final answer; it is the demand, arrival, workload, price, or failure signal that drives a decision model.

Textbook OR coverage often treats forecasting alongside inventory, capacity, simulation, and decision analysis because bad forecasts can make an otherwise correct optimization model recommend the wrong policy.

Core ideas

Time series

Time series methods use temporal patterns such as trend, seasonality, cycles, and autocorrelation to estimate future inputs.

Causal models

Causal or explanatory models connect forecasts to drivers such as price, promotions, weather, calendar effects, or policy changes.

Forecast error

Forecast error should be measured and carried into safety stock, staffing buffers, scenario design, and service-level calculations.

Scenario generation

Scenario generation converts point forecasts into plausible futures that stochastic, robust, and simulation models can evaluate.

Bias

Bias matters because a systematically high or low forecast creates recurring overstaffing, stockouts, excess inventory, or capacity shortages.

How to use it

  1. 1Identify which decision will consume the forecast and what time granularity it needs.
  2. 2Separate historical signal from calendar effects, interventions, outliers, and one-time shocks.
  3. 3Estimate both a central forecast and an error distribution or scenario set.
  4. 4Feed forecast uncertainty into the OR model rather than optimizing against a single point estimate.
  5. 5Backtest decisions, not only forecast accuracy, because the best statistical forecast is not always the best decision input.

Applications

InventoryStaffingRevenue managementCapacity planningEnergy
  • Inventory: forecast demand and error distributions to set replenishment quantities and safety stock.
  • Staffing: forecast arrivals and workload to choose shifts, breaks, and capacity buffers.
  • Revenue management: forecast demand by segment or fare class before controlling prices and availability.
  • Capacity planning: forecast long-run load before investing in machines, rooms, vehicles, or infrastructure.
  • Energy: forecast load, renewable output, and prices before dispatch and commitment decisions.

Common pitfalls

  • Optimizing against a point forecast while ignoring forecast error.
  • Judging the forecast only by statistical accuracy instead of downstream decision quality.
  • Training on historical data that no longer reflects pricing, policy, capacity, or customer behavior.
  • Using aggregate forecasts when the OR model needs disaggregated demand by location, time, product, or class.

Resources