# Portfolio Design

Advanced concepts in stock investment portfolio design.  Fundamentals, technical analysis and many other related topics are discussed.

# How Not to Optimize a Simulation

Today I would like to demonstrate one of the most basic of mistakes that can be made during optimization of a trading system. For background information please refer to Time The Stock Market Using Credit Spreads.

Stock analysis tools provided by Portfolio123.

The system for this demonstration is a simple market timer based on the BofA Merrill Lynch US Corporate BBB Option-Adjusted Spread index. I'm not going to bore the reader with the intimate details of how to set up the simulation and optimizer study as I will be keeping this post short. In summary, the simulation uses the Buy rule:

Eval(Close(0 ,##CORPBBBOAS) < (SMA(100, 0, ##CORPBBBOAS) + SMA(100, 100, ##CORPBBBOAS)) / 2, Ticker("RSP"), FALSE)

The Buy rule tests the closing price of the index against the 200 day Simple Moving Average (SMA) of the index.  I have split the moving average into two parts due to limitations imposed by Portfolio123 on the SMA parameters i.e. maximum period of 500, maximum offset of 500. The Buy rule is the starting point, and the Optimizer will successively increase the length of the moving average.

The permutations for the Optimizer study are set up as shown below, with the index price compared against the 200-day SMA, 400-day SMA, 600-day SMA, 800-day SMA, and 1000-day SMA. The last moving average is approximately four years, the longest that can be practically tested.

Optimizer permutations

The reason for testing for such a long moving average is that there is no "normal" spread that one can assume, as there is for the TED Spread (0.5%).  So the long average represents a nominal spread for recent history that is used as "normal".

The results of the Optimizer study are shown below. (Click to enlarge.)

Optimizer backtest results

Note that the return goes up with each permutation, and the drawdown goes down.  Naturally, I was drawn to the fifth permutation, the one with the highest return and close to the lowest drawdown.

This is where I stopped when I originally designed this market timer. However, that was an error, because the results were partially attributable to limitations in the simulation that I didn't investigate.

Below is a Portfolio123 multi-chart of the BofA Merrill Lynch US Corporate BBB Option-Adjusted Spread, from its start date, December 1996, or lets say January first 1997, as the chart appears to have a resolution of one month.