Quality Money Management : Process Engineering and Best Practices for Systematic Trading and Investment

258 CHAPTER◆ 2 8 Perform SPC Analysis

A common complaint about SPC is that it reacts too slowly. The real question is

the following: would a risk manager stop a trading/investment system sooner? And the

answer is no, for two reasons:

● A risk manager never stops a trading system when it makes too much money, even

though the predictive nature of the algorithm may be flawed and no longer predictive.

● SPC does not work only on three standard deviation control limits. A “ yes ” answer

to any of these questions may indicate that the system is out of control:

• Are points outside the upper or lower control limits?


• Do two out of three points land above or below the two standard deviation limits?


• Do four of five consecutive points fall above or below a one sigma limit?


• Do seven consecutive points fall either above or below the center line?


• Do seven or more points trend in the same direction?


• Do all points hug the center line, within plus or minus one standard deviation?^2

SPC measures means, standard deviations, and ranges of many different things, any one of

which may signal an out-of-control state. If anything, SPC may be too sensitive! Of course,

the more tests, the greater the probability of a false alarm. This can be controlled by smooth-

ing data through sampling. The thoroughness of SPC vastly exceeds the standard risk review

process at most firms.

With both a UCL and LCL, winning big is just as bad as losing big; in either case SPC

indicates that the process has changed. If the system is out of control to the upside, then we

may expect mean reversion or lack of future predictability. So, maybe the risk manager should

reduce the risk capital or suggest buying puts. The product team may need to retune the trade

selection algorithm for a new environment. Or, maybe traders choose to accept the risk, but

remain cognizant of the necessity to get out fast. Of course, any change must cause the prod-

uct team to revert to Stage 1 to ensure the change ’ s ability to add value. We also recommend

that the team run shadow trading as well, as these results can be the new benchmark.

28.3.2. d and p Charts

Manufacturing industries use several types of control charts for defective parts. Two types

of control charts are for the number of defective units, usually known as a d chart, and the

control chart for fraction defective, known as a p chart.

R control chart

0

0.01

0.02

0.03

0.04

0.05

0.06

0.07

0.0 8

R

UCL

CL

LCL

FIGURE 28-3