Quality Money Management : Process Engineering and Best Practices for Systematic Trading and Investment

The actual risk measurements can be displayed using quality control. We use these risk

calculations and SPC to determine whether the input process(es) to the trading/investment

algorithm has shifted. If the input process has changed, then we need to investigate the causes

of the change in the risk values. The cause could be a shift in the underlying stochastic proc-

ess caused by, say, economic or political issues, or a change in the regulatory environment.

Therefore, this spiral has been designed in an interleafed fashion. The first spiral is the

measurement of the algorithm against itself. This is classical risk management. The sec-

ond spiral is the measurement of the algorithm against a benchmark. The last spiral is the

measurement of the algorithm against the universe through simulation of the future.

The three categories of risk measurements are essentially snapshots at a point in time.

SPC links all these through time to find out what is normal, and to find changes in inputs

and processes early, not late. By addressing these items early, a trading/investment sys-

tem should be able to outperform its competitors who adjust after losses have taken place,

when everyone understands the risks in hindsight.

25.4. LOOP 1: Assess Single Performance

STEP 1: Specify Single Performance Controls

STEP 2: Benchmark Single Performance Calculations

STEP 3: Perform SPC Analysis

STEP 4: Determine Causes of Variation in Single Performance

In this step, risk managers assess the performance of a trading/investment system versus

itself, that is, versus the performance observed during the backtest. The first round of

risk control measures the system ’ s performance versus the out-of-sample test results from

Stage 2 to see if the machine is producing the expected performance metrics.

25.5. LOOP 2: Measure Performance Attribution

STEP 1: Define Benchmarks and Attribution Controls

STEP 2: Benchmark Attribution Calculations

STEP 3: Perform SPC on Attribution Metrics

STEP 4: Determine Causes of Variation in Attribution

Performance attribution will analyze performance of a trading/investment system relative

to its benchmark. We believe that risk managers should perform SPC on excess returns

using attribution analysis along with the over- and underweighting versus key factors. The

SPC will indicate if the machine is working according to the initial design constraints.

A stock-selection machine was severely overweighting in tech stocks in 1998 and 1999. This was justified

by excess returns. However, in 20 years of historical data, the algorithm never overweighted a sector to

this extent, nor were the excess returns ever to this level. Therefore, the machine was deemed to be out of

control. The owners had two choices: shut down the machine, or monitor the machine for early signs of

further breakdown and possible reversion to the mean. When the sector ’ s excess returns turned negative,

a human overrode the machine and brought the tech sector weight to below the benchmark weight. This

allowed the fund to outperform all other peer funds that remained overweighted in technology till 2001.

25.5. LOOP 2: MEASURE PERFORMANCE ATTRIBUTION 233