252 CHAPTER ◆ 2 8 Perform SPC Analysis
100% inspection of inputs and outputs since there are essentially no additional costs ver-
sus sampling. Nevertheless, sampling will remove noise from the data.
Applying statistical process control is contrary to the view of most financial mathe-
maticians, who are constantly looking for closed-form equations that accurately describe
process variation and predict the future. We use SPC to monitor the process and to
determine when a statistical process has changed. SPC does not try to forecast, rather, it
monitors. Flags raised by SPC spur investigation into the root cause of the change using
statistics and common sense.
The determination of the root cause of the change in the process in quality is normally
a blend between applied statistics and fundamental knowledge of the process. Perhaps it
is the separation between the scientists who believe they can forecast the future and the
practitioner who tries to explain using simple facts that has led to SPC not being adopted
in finance. What we are trying to accomplish with SPC in finance is to use SPC to iden-
tify when the underlying process, either inputs or outputs, have switched distributions.
We, like most people in finance, believe in mixtures of normals with jumps, which are
difficult (if not impossible) to forecast. The biggest objection to this form of mathematics
in finance is that it is not predictive, but simply descriptive of the past. This is a large step
forward from simpler risk limits. Now a trader or risk manager and top management can
make informed decisions about when to rely on the machine and when not to rely on the
machine. When the machine is broken, the firm will need a good quality engineer, who
combines statistics, common sense, and a deep understanding of the underlying proc-
esses, to oversee a kaizen team tasked with finding the root cause and proposing solutions
that might lead to a reformation of a product team.28.1. A Brief History of SPC
The history of statistical process control goes back to the 1920s and Western Electric,
part of Bell Labs, where Walter Shewhart perceived that real-world processes, such as
manufacturing, rarely generate normally distributed data. He concluded that these real-
world processes display variation that is inherent in the process (i.e., common variation)
and extra variation (i.e., special variation). The absence or presence of special variation
determines whether a process is in or out of control. Shewhart ’ s book The Economic
Control of Quality of Manufactured Product demonstrated that industrial processes could
yield data, which through statistical methods could signal the state of control of the pro c-
ess.^1 Deming applied Shewhart ’ s methods during World War II to improve the quality of
weapons. After the war he took SPC to Japan and transformed Japanese industry.
The underlying assumption in SPC is that any production process will produce prod-
ucts whose properties vary slightly from their designed values, even when the produc-
tion line is running normally, and these variances can be analyzed statistically to control
the process. For example, a machine may turn out parts 2 inches long, but some parts
will be slightly longer than 2 inches, and some will be slightly shorter, producing a
distribution of lengths. If the production process itself changes (say, e.g., the machine
breaks), this distribution can shift or spread out. If this change is allowed to continue,
the parts produced may fall outside the tolerances of the customer, causing product to be
rejected.
Quality Money Management should ensure that all the activities necessary to design,
develop, and implement a trading/investment system are effective and efficient with