Statistical methods of demand forecasting

This article provides a managerial overview rather than a complete treatment of statistical theory. Its purpose is to show where a few established techniques can make demand analysis more disciplined.

When to use it

• Use statistical methods when the HOOF approach would benefit from stronger evidence, particularly when estimating historical growth or testing the relationship between demand and an external driver.

Origins

The techniques come from different statistical and economic traditions rather than one unified model. Least-squares regression developed through the work of Adrien-Marie Legendre and Carl Friedrich Gauss; trend extrapolation grew with time-series analysis; and the National Bureau of Economic Research developed indicator-based, or barometric, methods through business-cycle research led by Wesley Clair Mitchell, Arthur Burns and others.

What it is

The familiar warning about “lies, damned lies and statistics,” often associated with Mark Twain, is a reminder to use evidence critically rather than a reason to avoid quantitative analysis. Three comparatively accessible methods can support market-demand work:

Trend projection

Regression analysis

Barometric method (NBER)

How to use it

Trend projection

A trend line summarises the direction of observations plotted over time:

Set out annual market-demand data.

Plot time on the x-axis and demand on the y-axis, using logarithmic graph paper when a constant percentage growth rate is the relevant pattern.

Place a reasoned line of best fit through the observations.

Measure its gradient to estimate the average annual growth rate.

This provides an alternative to the moving-average approach in Appendix A. Both methods should produce a broadly comparable estimate of historical market growth when the assumptions fit.

Treat extrapolation with much greater caution. Extending the line across future periods produces numbers, but not necessarily credible forecasts. The projection assumes that the forces shaping past demand will continue in the same way. That is often unrealistic. Use the line primarily to understand history, then apply the HOOF approach to demand forecasting to examine how demand drivers may change.

Regression analysis

Regression estimates how a dependent variable such as market demand varies with an independent variable such as GDP or engineering output. For the simple linear case shown in Figure C.1:

Set out demand for each period.

Record the chosen independent variable for the same periods.

Plot the independent variable on the x-axis and demand on the y-axis.

Fit a straight line through the points.

Estimate its slope, m, and the y-axis intercept, c, when x = 0. The fitted relationship takes the standard form y = mx + c.

Figure C.1 Regression analysis

c

Independent variable

A historical association may not persist, and it does not by itself demonstrate causality. Other drivers can become more important in the future, while the independent variable may be jointly influenced by the same conditions as demand. Use regression to quantify a past relationship and test uncertainty, then combine it with causal reasoning and the HOOF process before forecasting.

Barometric method (NBER)

The National Bureau of Economic Research developed a composite-indicator approach to economic forecasting. Apply it to market demand as follows:

Identify economic indicators with a plausible influence on demand.

Convert each indicator into a comparable time series of index values.

Assign weights reflecting the evidence on relative influence.

Combine the weighted series into a composite index.

Compare the historical composite with an index of past market demand.

Forecast the component indicators and composite.

Translate the projected composite into a demand forecast.

Timing can be included by classifying indicators as:

Leading – variables that tend to move before demand, such as new orders before recorded sales.

Coincident – variables that move broadly alongside current economic activity.

Lagging – variables that respond after a delay, such as short-term lending following a change in interest rates.

The barometric method is more elaborate than a simple trend line or single-variable regression and captures part of HOOF’s driver-based logic. It also has important limits. Users may struggle with indices, weights, lags and aggregation; future weights and delays may differ from historical ones; and qualitative forces such as awareness or fashion may resist credible conversion into an index.

Top practical tip

Begin with a causal question and a clean historical series. A sophisticated calculation cannot rescue an incoherent market boundary or unreliable input data.

Top pitfall

Do not assume that a past trend, regression or indicator relationship will continue. State the mechanism, test alternative drivers and make structural change explicit.

Further reading

• Hyndman, R.J. and Athanasopoulos, G. (twenty twenty-one). Forecasting: Principles and Practice. OTexts.
• Box, G.E.P., Jenkins, G.M., Reinsel, G.C. and Ljung, G.M. (twenty fifteen). Time Series Analysis: Forecasting and Control. Wiley.