254 CHAPTER ◆ 2 8 Perform SPC Analysis
and inputs have changed. Therefore, you must investigate the root cause to determine
the special cause. In current methods of risk analysis, there are two outcomes: one, the
system performs perfectly, or two, it performs imperfectly. The impulse is to react to
imperfect behavior with modifications to the system without determining if the variation
is normal or abnormal.
With quality, action is only taken when the trading/investment system is out of con-
trol. If control limits are hit, or patterns inside the limits are experienced, then and only
then is action taken. What we are proposing is the use of statistical process control, which
triggers no changes as long as a trading/investment system is in control. If there are
exceptional returns, good or bad, then the system is out of control. For example, if the
top ranked stocks underperform the bottom ranked stocks more than usual, the system
is out of control. Does this mean that the system must be down? Maybe. Does this mean
switching to manual trading from automatic trading? Probably. Does this mean start look-
ing for assignable, root causes? Absolutely. SPC charts:● Figure out a special model to get you through the shift, all along still using the
original model.
● Figure out the root cause.
● Remove the special model and return to the original once stability comes back.Risk calculations and reports give management a snapshot of the returns and potential
losses and drawdowns both on an absolute basis and relative to the benchmark over a
given time horizon. Essentially, these reports will help us to understand how the system is
performing relative to the market and the rest of the industry.
The actual application of SPC depends on the type of trading/investment and the
backtesting method. Metrics and their SPC charts will be different for every system. The
standard application of SPC is to graphically monitor the trading/investment system out-
puts at regular intervals. Maybe every day.28.3. SPC Control Charts
The purpose of statistical process control for trading is a new concept for most money
managers. The basis of this concept is on the following facts:● A trading algorithm ’ s ability to produce excess returns is stochastic since the under-
lying process is stochastic.
● A trading algorithm ’ s measurable output will be stochastic because the underlying
process is stochastic.
● Most quants try to build trading algorithms that produce an output with zero varia-
tion. Many quants believe they have achieved this goal until their algorithm blows
up and loses money as described in Teleb ’ s book Fooled by Randomness.
● SPC monitoring of a trading system based on a stochastic process has two key
outputs:- The mean of the input and/or outputs.
- The standard deviation of the inputs and/or outputs.
● A trading system based on a stochastic process needs a tool to determine if the out-
put is exhibiting normal variation or if the underlying process has changed.