255
The tool that has been developed over the last 80 years for the above facts is called SPC. The
goal of this tool is not to predict the future, but to notify the operator of the machines that
the process is no longer deemed to be stable and in control.
A control chart is a record of the performance outputs, a running record of the system.
Control limits are boundaries on a control chart within which a trading/investment system
can safely operate. These limits are based on past performance (i.e., backtest) and show
what to expect from the system when nothing has changed. Each time a risk manager
checks the system, he should compare the output results with the control limits. If the
results are within the limits, then there is no problem. If some points on a control chart
fall outside the control limits, something has happened and the trading/investment system
is no longer operating normally. There are two types of control charts:● Variables charts, which are used where a performance metric is a number. The
X bar and R charts are variables charts.
● Attribute charts, which are used where a performance metric is Boolean, either
acceptable or unacceptable, like say winning or losing trades. Quality practitioners
assign numerical values to the results to enable statistical analysis. A percent defec-
tive chart for winning trades versus losing trades is an example of a variables chart.
These are particularly helpful for monitoring trigger trading systems.The enactment of Sarbanes-Oxley is a perfect example to illustrate the benefits of
SPC. Sarbanes-Oxley changed the validity of historical earnings data and research. When
the bill became law, every firm that used earnings per share as an indicator needed to
completely retune their algorithms, because the input process changed. While it is now
well understood that this change shifted stock volatility around earnings, index volatil-
ity during key earnings months, it was not understood at the time. By applying SPC on
outputs, a risk manager could have identified the shift. Analysis would have identified the
root cause. A product team could have retooled and backtested again.
A firm may need to build parallel systems, ready to go if there is a regime
switch. This is what keeps a firm at the top. The replacement technique is far superior
to the “ I got caught in a regime switch and lost faith in the model so now I am
making side bets out of control to keep my job ” technique. SPC will detect early shifts
in assumptions in models, shifts in distributions, and forces risk managers to look at
and understand what has changed in the market. SPC forces statistical decisions about
trading strategies. The amount of money a trader can make by successfully driving
around a perceived short-term shift is pennies on the dollar compared to what a system
can make by understanding the long-term, fundamental nature of the shift and responding
correctly.28.3.1. X bar and R Charts
X bar and R charts are the most commonly used SPC charts. Together X bar and R moni-
tor the behavior of a process over time. X bar shows sample means, and R sample ranges.
Interpreted together they allow tracking of both the center and variation of a pro cess,
and detection of out-of-control performance. To illustrate the calculations for X bar and
R charts we will use the following table of 20 weeks ’ worth of daily returns for a trading
system.28.3. SPC CONTROL CHARTS