Unit 3: Demand Estimation

Demand estimation is the quantitative and judgemental exercise of measuring the current level of demand for a product and projecting its future level, so that a firm can plan output, pricing, inventory and investment. It sits between demand theory (which explains why quantity demanded responds to price, income and related variables) and managerial decision-making (which needs actual numbers). Estimation looks at present or historical demand; forecasting extends the estimate forward in time.

Defining features that later sections rely on:

• Object measured: the quantity demanded (Qd) as a function of determinants — own price (P), income (Y), price of related goods (Pr), tastes, and time.
• Distinction: estimation fits a relationship to existing data; forecasting predicts future values from it.
• Data types: time-series (one variable observed over successive periods) versus cross-section (many units at one point in time).
• Approaches: qualitative (opinion- and judgement-based, used when data are scarce) versus quantitative (statistical, needing historical numbers).
• Accuracy criterion: a forecast is judged by its error — the gap between forecast and actual, often summarised as Mean Absolute Deviation (MAD) or Mean Squared Error (MSE).
• Time horizon: short-run (up to 1 year, operational), medium-run (1–3 years), long-run (over 3 years, strategic).

II. Relevance of Demand Estimation for a Firm

Why measuring and predicting demand underpins nearly every managerial decision.

A. Purpose and principle

Demand estimation converts market uncertainty into planning figures, letting a firm commit resources with a defensible expectation of sales.

• Reduces uncertainty: replaces guesswork with a numeric expected demand, e.g. projecting 50,000 units next quarter rather than "probably a lot".
• Links to elasticity: an estimated demand function yields price elasticity (Ep), guiding whether a price cut raises revenue. If Ep = −2, a 10% price cut raises quantity 20% and revenue rises.

B. Relevance of demand estimation for a firm

The concrete decisions that depend on a reliable demand estimate span the whole firm.

• Production planning: sets the output schedule and capacity utilisation; a forecast of 12,000 units/month fixes shift patterns and raw-material orders.
• Pricing decisions: an estimated demand curve shows the revenue-maximising price where marginal revenue (MR) = 0, and supports price discrimination across segments.
• Inventory control: forecast demand feeds reorder points and safety stock, avoiding both stock-outs and carrying-cost waste.
• Investment and capacity: long-run forecasts justify a new plant or line; overestimation leaves idle capacity, underestimation forfeits sales to rivals.
• Financial planning: projected sales revenue underlies cash-flow budgets, working-capital needs and loan servicing.
• Marketing and promotion: estimating advertising elasticity of demand shows the payoff of a campaign; sales targets and territory quotas derive from forecasts.
• Manpower planning: expected volume determines hiring, training and overtime budgets.
• New-product launch: even without history, estimation gauges likely uptake so launch scale and distribution match expected demand.

III. Demand Forecasting Using Qualitative Forecast

Judgement-based methods used when historical data are absent, unreliable, or the environment is changing too fast for numbers alone.

A. Nature and conditions

Qualitative (or subjective) forecasting relies on informed opinion rather than mathematical extrapolation, and dominates for new products, long-run strategy and turbulent markets.

• When used: no time-series exists (new launch), or structural change (new technology, regulation) makes past data misleading.
• Basis: expert intuition, buyer intentions and field experience, aggregated systematically.
• Trade-off: flexible and forward-looking, but prone to personal bias and hard to replicate.

B. Qualitative forecast methods

Each method gathers and pools judgement from a different source.

1. Survey / opinion-based methods — collect intentions directly.

   • Consumer survey (buyer's intention): ask a sample of buyers how much they intend to purchase; scale up to the population. Good for short-run, but stated intent often exceeds actual purchase.
   • Sales-force opinion (composite): each salesperson estimates demand in their territory; the firm sums territorial figures. Close to the market, but reps may understate to win easier quotas.
   • Expert opinion / executive jury: senior managers from finance, marketing and production pool their forecasts into a consensus figure. Fast and cheap, but dominated by strong personalities.

2. Structured / group methods — reduce bias through process design.

   • Delphi technique: a panel of experts answers in successive anonymous rounds; a coordinator feeds back the summary of each round, and experts revise until answers converge. Anonymity removes bandwagon and dominance effects.
   • Market experiment / test marketing: the product is sold in a limited region or under varied prices, and the observed response is projected to the whole market. Real behaviour, but costly and it alerts competitors.

C. Applications and limitations

Qualitative forecasts earn their place precisely where data-driven methods cannot start.

• Best fit: new products, long-run strategic questions, and demand for capital goods where buyers are few and identifiable.
• Strengths: incorporate expected future events (a coming law, a rival's exit) that no historical series contains.
• Limitations: subjectivity and bias, non-repeatability, and difficulty of expressing confidence numerically; often combined with quantitative methods for cross-checking.

IV. Demand Forecasting Using Time Series Analysis

Projecting future demand by decomposing a historical sequence of sales into its regular components and extending them.

A. Definition and components

Time series analysis assumes the future is an extension of the past, so it isolates the systematic patterns in a variable observed at equal time intervals and carries them forward.

The observed value Y in any period is treated as a combination of four components:

Multiplicative model:  Y = T × S × C × I
Additive model:        Y = T + S + C + I

• T — Trend: the long-run upward or downward movement (e.g. rising smartphone sales over a decade).
• S — Seasonal: regular fluctuations within a year (umbrella sales peak in monsoon).
• C — Cyclical: wave-like swings over several years tied to the business cycle (boom and recession).
• I — Irregular: random, unpredictable shocks (a strike, a flood).

Where Y = value of the series, and T, S, C, I are the trend, seasonal, cyclical and irregular components.

B. Time series analysis methods

The techniques differ in how they extract the trend and how much weight they give to recent versus older observations.

1. Smoothing methods — average out irregular noise.

   • Simple moving average (SMA): the forecast for the next period is the mean of the last n actual values.
     
     TEXT

          F(t+1) = ( Y(t) + Y(t-1) + ... + Y(t-n+1) ) / n

     
     where F = forecast, Y = actual, n = number of periods averaged. A larger n smooths more but lags turning points.
   • Weighted / exponential smoothing: recent periods get more weight.
     
     TEXT

          F(t+1) = α·Y(t) + (1 − α)·F(t)

     
     where α is the smoothing constant (0 < α < 1). High α reacts fast to change; low α gives a stable forecast.

2. Trend-projection methods — fit and extend a systematic trend.

   • Least-squares linear trend: fit a straight line minimising the squared errors.
     
     TEXT

          Y = a + b·X
          b = Σ(XY) / Σ(X²)      a = ȲŸ (mean of Y)   [X coded so ΣX = 0]

     
     where X = coded time, a = intercept (value at X = 0), b = slope (change in Y per period).
   • Seasonal index method: compute a ratio-to-moving-average for each season, then multiply the projected trend by the index to restore seasonal shape.

C. Worked example — least-squares trend

Fitting a trend to five years of sales shows how the projection is produced.

• Data (units, '000): Year 1–5 sales = 10, 12, 13, 16, 19.
• Code time so ΣX = 0: X = −2, −1, 0, 1, 2; mean sales Ȳ = 70/5 = 14.
• Compute slope and intercept:
  
  TEXT

    Σ(XY) = (−2·10)+(−1·12)+(0·13)+(1·16)+(2·19) = 22
    Σ(X²) = 4+1+0+1+4 = 10
    b = 22 / 10 = 2.2      a = 14

• Trend line: Y = 14 + 2.2X. For Year 6, X = 3, so forecast = 14 + 2.2(3) = 20.6 thousand units.

D. Applications and limitations

Time series methods suit stable, data-rich, short-to-medium-run problems.

• Best fit: established products with several years of consistent history and no structural break.
• Strengths: objective, repeatable, cheap once data exist, and able to separate seasonality from trend for operational planning.
• Limitations: assumes past patterns persist, so it misses turning points caused by new events; ignores causal variables like price or income; the irregular component (I) is by definition unpredictable, capping accuracy.
• Complement: firms often pair a time-series baseline with qualitative adjustment to fold in expected future changes the data cannot show.