232 CHAPTER ◆ 2 5 STAGE 4: Overview
Furthermore, systems are not scaled up or down in backtests; they assume a steady state.
The snapshot view of risk does not distinguish between what is a normal, or common
cause, variation and what is a special, or assignable cause, variation. The snapshot view
of risk ignores tolerancing and, furthermore, lacks a stochastic view of risk outputs aris-
ing from stochastic inputs from the real world.
25.2. A Brief History of Quality Control
In industrial engineering, engineers learned a long time ago that a stochastic process can
be controlled using very basic statistical tools. The tools do not fix a broken process but
tell engineers when a process is no longer working as designed. In the K | V methodol-
ogy we apply this quality process control to manage the automated trading/investment
systems. The key point of SPC is to use statistics to quickly identify when a process
mean, standard deviation, or range has shifted due to special causes. In manufacturing we
are able to measure most of the inputs and outputs of the process. In trading/investment,
we are not able to measure many of the inputs into the system since we do not know for
sure that we are in a recession until we are nine months into a recession.
It is well understood in quality that an engineer should not adjust a machine after every
part, which leads to overcompensating. In SPC as long as parts are within tolerance, that
is, within control limits, no adjustments are made, a process which produces higher qual-
ity parts. The analogy to finance is traders lowering risk capital during periods of draw-
downs, which leads to being undercapitalized when the market bounces back. When it
does bounce back, they often overcompensate before a big downturn.
One can perceive dollar cost averaging as a quality view of the market. During a down
market, dollar cost averaging does not consider the machine broken. The market is not
broken, so keep adding money to the market machine. No changes are made to invest-
ment policy in either up or down markets. The CBOE S & P 500 BuyWrite Index (BXM)
is also a quality control view. (A buy/write strategy is one where the investor buys a stock
(or a stock index) and writes call options (i.e., the options are “ covered ” ) on the stock
(or stock index).) Option premium is highly correlated with market uncertainty. When
investors are uneasy, they demand more premium. When they are sure, they demand less.
Effectively, the market widens its tolerance level in uncertain times. When the market
goes down, an investor gets paid for the increased uneasiness by selling higher premiums.
The BXM is like trading an SPC mechanism. When control limits are hit, it naturally
changes the trading structure by selling volatility.
25.3. Combining Performance and Quality Control
The combination of classical risk calculations and the application of quality control for
automated trading system are based on our beliefs that a trading algorithm works in a
stochastic environment. The algorithm does not remove a significant portion of the
stochastic nature of the environment and thus the output of these algorithms is stochas-
tic. The standard control of the stochastic process embedded in the output is to build a
large portfolio. The large portfolio is used to create a stable mean, based on the law of
large numbers. The stable mean concept is identical to statistical quality control where
we take a sample and measure the mean, standard deviation, range, etc. The distribution
of these sample means is normal due to the central limit theorem. The distribution of the
sample means is then used to control the process through SPC charts.