Future Savvy: Identifying Trends to Make Better Decisions, Manage Uncertainty, and Profit from Change

CHAPTER 7

The Limits

of Quantitative

Forecasting

This page intentionally left blank

155

I

N THE PREVIOUS CHAPTER WE DISCUSSED PROBLEMS

with trend extrapolation as an approach to future

prediction. But we dealt essentially with problems

in single-issue projection. Is it not perhaps possible

to consider the simultaneous interaction of many

variables, perhaps with a complex statistical model

backed by computer power, and so achieve a trend

projection we can rely on? In this chapter we con-

sider where this is possible and useful, and where

it is not.

Statistical Analysis and Quantitative Modeling

Our lives and decisions are dominated by statistics. For better or

worse, quantitative analysis has become the authoritative form of

knowledge. In fact, this is a relatively new phenomenon in the his-

tory of human thought. It is only since the nineteenth century that

statistical analysis has come to stand at the core of the way we

think about the world, and alternative forms of investigation—judg-

ment, experience, and intuition—have been pushed into the back-

ground. But the pendulum has swung so far that, in our era, not

only have quantitative approaches become central to how we in-

vestigate complex situations, but also unless something is numer-

ically studied, it is almost “not knowledge.” Economics, once an

arena of social analysis, has become a field of turbo-math, while

management academics produce papers that more closely resem-

ble particle physics than anything real managers actually do. This

pattern is repeated across much of psychology and the rest of the

social sciences.

Ready access to computer power, allowing us to do more with

numbers, has greatly facilitated this shift. So it should come as no

surprise that many people look to sta-

tistical analysis or quantitative meth-

ods, and particularly computer-driven

projective modeling, to solve the

conundrum of predicting the future.

Software developers and entrepre-

neurs have taken up the challenge to

develop computer-driven forecasting

methodologies with alacrity. Future-

oriented number-crunching software programs, with such names

as Autocast, ForecastX, Forecast Pro, and SmartForecasts, are often

lavishly advertised to corporations and other institutions that have

156

✧

FUTURE SAVVY

So it should come as no sur-

prise that many people look

to statistical analysis or

quantitative methods, and

particularly computer-driven

projective modeling, to solve

the conundrum of predicting

the future.

an interest in anticipating and evaluating change. Forecast Pro’s

advertising text is typical of the kind: “With Forecast Pro, you pro-

vide the historic data for the items you are forecasting and Fore-

cast Pro does the rest. The built-in expert selection mode analyzes

your data, selects the appropriate forecasting technique and calcu-

lates the forecasts using proven statistical methods.”

1

All computer-driven quantitative modeling generates future pro-

jections through predictive algorithms, based on mathematical re-

lationships between variables derived from analysis of past data.

Patterns and associations among variables and multiple relation-

ships of cause and effect (causal and dependent variables) may be

derived, for example, by regression techniques, to determine what

mix of causal influences are at work on any perceived outcome

variable, and the degree of influence of each on the outcome. (A

slump in sales may be attributed to many factors: less advertising,

placement of outlets, lower consumer spending, competitive influ-

ences, distributor discounts, product options, changes in house-

hold income, etc. A regression can help determine what mix of

causal factors in what proportion has caused this over time.)

Modeling techniques, including econometrics, set the deduced

coefficients between variables into complex algorithms that math-

ematically define the relationships between variables and set deci-

sion points about which variables affect which others, under what

conditions, and by how much. With this, the modeler can project

into future time, allowing us to watch the future evolving on a

computer screen.

The most common form of quantitative projection is based on a

“time-series analysis,” where a sequence of data points measured

at successive and preferably uniform time intervals has been col-

lected. Using the standard measure of time as a base, analysts cre-

ate a mathematical curve that approximates the data evolution to the

CHAPTER 7: THE LIMITS OF QUANTITATIVE FORECASTING

✧

157