Forecasting & Aggregate Production Planning

Forecasting is a compact GATE ME topic (~1–2 marks). The moving-average and exponential-smoothing formulas plus error measures are the reliably tested items.

Forecasting methods

• Simple moving average (n periods): F = average of the last n demands. Larger n → smoother but slower to react.
• Weighted moving average: recent periods get higher weights (weights sum to 1).
• Exponential smoothing: F_{t+1} = F_t + α(D_t − F_t) = α·D_t + (1 − α)·F_t, where α (0–1) is the smoothing constant. Larger α → more responsive (weights recent data more); smaller α → smoother.
• Trend/seasonality: double exponential smoothing (Holt) for trend; linear regression for a trend line.

Forecast error measures

• MAD = mean absolute deviation = Σ|D − F|/n.
• MSE = mean squared error = Σ(D − F)²/n (penalises large errors).
• MAPE = mean absolute percentage error. Bias/tracking signal detects consistent over/under-forecasting.

Aggregate planning

Match capacity to demand over a medium horizon using level (constant output, vary inventory), chase (match output to demand, vary workforce), or mixed strategies — trading inventory, hiring/firing, overtime, and backorder costs.

Exam Tricks & Tips

• 🎯 Exponential smoothing F_{t+1} = α·D_t + (1 − α)·F_t — the single most-tested forecasting formula.
• 🎯 Larger α = more responsive (follows recent demand); smaller α = smoother/more stable — pick per demand volatility.
• 🎯 Exponential smoothing weights decay geometrically into the past — every past period contributes, unlike a fixed-window moving average.
• 🎯 MAD averages absolute errors; MSE squares them (punishes big misses) — choose the right accuracy metric.
• 🎯 Level strategy uses inventory to buffer; chase strategy varies the workforce — the two aggregate-planning poles.
• ❌ Common mistake: misreading α — a high α does NOT mean more smoothing; high α makes the forecast MORE reactive (less smoothing).

Expected exam pattern

A 1-mark exponential-smoothing or moving-average forecast NAT, and a 2-mark forecast-error (MAD/MSE) or next-period forecast problem. The role of α and the level-vs-chase distinction appear as MCQs.

Quick recap

Moving average (n periods, larger n smoother); weighted MA; exponential smoothing F_{t+1} = αD_t + (1−α)F_t (high α = responsive). Errors: MAD = Σ|D−F|/n, MSE = Σ(D−F)²/n. Aggregate planning: level (buffer inventory) vs chase (vary workforce).

Cover the answer, recall, then check. 11 cards on forecasting for GATE ME.

Q1. Formula for an n-period simple moving average forecast?
A1. F = (sum of the last n demands)/n. Larger n gives a smoother, slower-reacting forecast.

Q2. State the exponential smoothing formula.
A2. F_{t+1} = F_t + α(D_t − F_t) = α·D_t + (1 − α)·F_t, with smoothing constant α ∈ [0,1].

Q3. Effect of a large vs small smoothing constant α?
A3. Large α → more responsive to recent demand (less smoothing); small α → smoother, more stable forecast.

Q4. How do past demands contribute in exponential smoothing?
A4. Their weights decay geometrically as α(1−α)^k — every past period contributes, with older ones weighted less.

Q5. Define MAD.
A5. Mean Absolute Deviation = Σ|D − F|/n — the average magnitude of forecast error.

Q6. Define MSE and its property.
A6. Mean Squared Error = Σ(D − F)²/n; it penalises large errors more heavily (squares them).

Q7. What is a weighted moving average?
A7. A moving average giving higher weights to more recent periods; the weights must sum to 1.

Q8. Which method handles a demand trend?
A8. Double exponential smoothing (Holt's method) or linear regression, which add a trend component.

Q9. Difference between level and chase aggregate strategies?
A9. Level: constant output, absorb demand swings with inventory. Chase: vary output/workforce to match demand each period.

Q10. What is a tracking signal used for?
A10. To detect persistent bias (consistent over- or under-forecasting) by comparing cumulative error to MAD.

Q11. Common misconception about α?
A11. That high α means more smoothing — actually high α makes the forecast more reactive; low α gives more smoothing.

Production planning starts with a demand forecast, then decides how to meet it economically over the medium term. GATE tests the moving-average and exponential-smoothing forecasting methods and forecast-error measures. These are practical, formula-driven numericals.

Core concept: forecasting projects future demand from past data; aggregate planning then sets output, workforce and inventory levels to satisfy that demand at least cost.

Deep explanation

Beginner — time-series forecasting

• Simple moving average (SMA): forecast = average of the last n periods. Larger n smooths more but lags more.
• Weighted moving average: recent periods weighted more heavily.
• Exponential smoothing: F_t = F_(t−1) + α(D_(t−1) − F_(t−1)) = α D_(t−1) + (1 − α)F_(t−1), where α (0–1) is the smoothing constant. High α responds fast to change; low α smooths noise.

Intermediate — trend, seasonality and error measures

• Data may have level, trend (use double exponential smoothing / Holt's method) and seasonality (seasonal indices). Decompose accordingly.
• Forecast accuracy measures:
  • MAD (mean absolute deviation) = Σ|error|/n,
  • MSE (mean squared error) = Σerror²/n,
  • MAPE (mean absolute percentage error) = (Σ|error/actual|/n)×100,
  • Bias/tracking signal = Σerror/MAD to detect systematic drift.

Advanced — aggregate production planning (APP)

APP balances demand over months/quarters using strategies:

• Chase strategy: vary output/workforce to match demand (low inventory, high hiring/firing cost).
• Level strategy: constant output, absorb variation with inventory/backorders (stable workforce, high inventory cost).
• Mixed strategies combine both. Costs traded off: regular/overtime, hiring/firing, inventory holding, backorder, subcontracting. The master production schedule (MPS) and MRP (materials requirement planning, exploding the bill of materials against lead times) turn the aggregate plan into detailed orders.

Worked example

Last month's forecast was 100 units, actual demand was 120 units. With exponential smoothing α = 0.2, find this month's forecast.
F_t = F_(t−1) + α(D_(t−1) − F_(t−1)) = 100 + 0.2(120 − 100) = 100 + 0.2(20) = 100 + 4 = 104 units.
(Only 20% of the error is corrected; a higher α would react more strongly toward 120.)

GATE relevance

Moving-average and exponential-smoothing forecasts, the smoothing-constant effect, forecast-error measures (MAD/MSE), and aggregate-planning chase-vs-level strategies plus MRP are recurring Industrial Engineering questions. The exponential-smoothing update formula is a guaranteed numerical.

Exam tricks & shortcuts

• Exponential smoothing: F_t = αD_(t−1) + (1−α)F_(t−1) — new forecast blends last demand and last forecast.
• High α → responsive; low α → stable/smooth.
• Chase = match demand (inventory-light); Level = steady output (inventory-heavy).
• Mnemonic: "Alpha weights the news, one-minus-alpha keeps the old view."

Key formulas

• Simple moving average: F_t = (ΣD of last n periods)/n.
• Weighted moving average: Σ(wᵢDᵢ).
• Exponential smoothing: F_t = F_{t−1} + α(D_{t−1} − F_{t−1}), 0 < α < 1.
• Trend (double smoothing/regression): y = a + bx.
• Error measures: MAD = Σ|error|/n; MSE = Σerror²/n; bias (running sum of errors).

Forecasting blends recent and past demand; MAD/MSE measure accuracy for choosing a model.

Worked Example

Problem: (a) Using single exponential smoothing with α = 0.2, the forecast for the current period was 100 units while actual demand turned out to be 120 units. Find the forecast for the next period. (b) If the demands of the last three months were 100, 120, and 110 units, find next month's forecast using a 3-month moving average.

Solution:
(a) Exponential smoothing update:
F_(t+1) = F_t + α·(A_t − F_t),
where A_t is actual demand and F_t is the old forecast.
F_(t+1) = 100 + 0.2 × (120 − 100) = 100 + 0.2 × 20 = 100 + 4 = 104 units.

(b) Simple 3-month moving average:
F = (100 + 120 + 110)/3 = 330/3 = 110 units.

The two methods differ because exponential smoothing weights the most recent error, while the moving average weights the last three periods equally.

Answer: (a) Next-period forecast = 104 units; (b) 3-month moving average forecast = 110 units.