269
Just as in manufacturing the only way to find these anomalies is to apply SPC theory
to results of the process. Once the process becomes out of control then a cause of the
out-of-control condition can be found. If we can find the cause of the process being out
of control then we can fix the process and have theoretically less variance than the bench-
mark. Sometimes in finance all we can do is identify the out-of-control state since we
cannot determine what has caused the algorithm not to perform properly. This is also use-
ful since we then can adjust our bets accordingly.
A close friend of ours, Peter Krause, discussed with us a unique anomaly. A high fre-
quency trading system ’ s win-to-loss ratio dropped in half for a single stock in one week.
The next week the ratio dropped for ten stocks. Then it dropped for 100 stocks. Finally,
the ratio dropped for all the stocks in the SPX and QQQ, which a friend was trading. Half
of the profit disappeared from the trading system within two months. A competitor had
turned on a duplicate system. The interesting point our friend made, however, was that
his friend ’ s firm did not foresee what was going on; they only knew they were making
less money. Only after investigating the data did they see the pattern, but by then it was
too late. It would have been nice to know ahead of time so the firm could have defended
its algorithm by actively driving the other firm ’ s test results through active trading.
This is why we recommend the use of SPC to control the algorithms. Had the firm
produced SPC charts for win/loss ratio by stock and by sector, they would have seen
the shift in the underlying process. The firm should have already developed a strategy
that would drive the profit in these first stocks to zero or negative. The firm should have
implemented this strategy quickly in the hopes the other firm would be confused by the
results and go back to the drawing board. If this failed, the firm should have had a new
system built and waiting to be implemented when a competing algorithm was placed into
production. However, without monitoring the process and using kaizen to improve the
process, the firm was stuck with half the profit until the next system was turned on, then
one-third the profit, then one-quarter the profit until the business becomes a low cost busi-
ness. If a firm is not the low cost trader, it will be driven out of the marketplace, not due
to the trading/investment strategy, but due to the fact that its profit margin has been cut.
While new in financial markets, this view is consistent with the new Lanchester strat-
egy (discussed in Chapter 4), which was originally developed in 1916—constantly looking
for changes in the environment, including new machines coming online. Do not rule out
new hardware to fight the war until the war is on. This is widely known by the U.S. mili-
tary. Use the weaponry needed to win the war. Always keep another layer of new technol-
ogy ready to go. SPC will tell you when the competition has rolled out a new machine.29.10. LOOP 3: Determine Causes of Variation in VaR
We also recommend the use of DoE and ANOVA and what-if scenarios to investigate and
determine causes of variation in Value at Risk. Monte Carlo simulation for Value at Risk
will provide the most flexibility for what-if scenarios.
A major benefit of building out a VaR system is the ability to perform what-if analysis.
Ideally, VaR identifies the securities, sectors, credit ratings, etc. that result in the greatest
shocks to the portfolio. The goal is to use the information from VaR to construct hedges.
The goal of hedging is to make returns more stable and the reduce downside deviation. If
properly built and implemented, a firm should be able to significantly reduce drawdowns
and increase returns, and therefore limit redemptions during market shocks. In a perfect
world, a firm can profit during market meltdowns and beat the benchmark during normal
markets by using well-constructed hedges and superior trading/investment systems.29.10. LOOP 3: CAUSES OF VARIATION IN VaR