261
A process with a low C pk should be thought of as a process that will require constant
monitoring.28.4. LOOP 1: Perform SPC Analysis on Single Metrics
All inputs and single factor ouputs should be monitored. The monitoring of these
signals allow an operator to determine if a fundamental shift in the predictive nature
of the factor has occurred. When an out-of-control situation occurs, the team needs
to focus on a quality process method such as Ford 8D to first contain and then fix the
root cause.28.5. LOOP 2: Perform SPC on Attribution Metrics
The output of attribution is normally considered to be stationary and nonstochastic.
We believe the output to be stochastic and should be monitored with SPC. The number
of standard deviations that an algorithm is over or under weighting a sector is mean-
ingful information. Even more meaningful is whether the over/under weighting is a
normal occurance or an out-of-control signal. These results should be fed back into the
kaizen system.28.6. LOOP 3: Perform SPC on VaR Metrics
The output of the VaR calculations, day to day, are stochastic. We recommend the use of
SPC to analyze and control VaR outputs. An example of this is the 95% loss threshold.
If, for example, seven days in a row, the VaR number has crept up, SPC would say the
underlying process mean is shifting, so risk managers should look at the machine to find
root causes. In traditional VaR, risk managers would otherwise wait until the VaR number
hits some prespecified limit.
If you take the kaizen approach, then VaR allows a risk manager attributes or vari-
ables, the largest influencers of the overall process variation, which need either model
enhancements, better hedging strategies, or a kaizen team to figure out the root cause.28.7. Summary
The concept of SPC for process monitoring is very old. The idea of monitor-
ing a machine, person, or algorithm is common in everyday life. Using SPC charts to
monitor and control a trading/investment process has not, however, been openly discussed
in finance.
We recommend all three types of risk outputs be monitored using SPC. We also recom-
mend risk managers use SPC to monitor and adjust trading limits in a scientific manner;
and we recommend the use of SPC to identify problem areas, as a start to the kaizen
process. Finally, we recommend that the trading/investment industry follow the manu-
facturing world by aggressively rooting out abnormal sources of variation in the perfor-
mance of trading/investment systems. We expect the removal of variation to produce even28.7. SUMMARY