Dangers of Compounding

Discussions about the testing and simulation of mechanical trading systems using historical data and other methods. Trading Blox Customers should post Trading Blox specific questions in the Customer Support forum.
Post Reply
TK
Roundtable Knight
Roundtable Knight
Posts: 167
Joined: Tue Apr 15, 2003 5:45 pm

Dangers of Compounding

Post by TK » Tue Sep 07, 2004 7:34 am

I have found an interesting post about how we can end up with curve-fitted portfolios when using compounding in tests and why we should consider using fixed position sizes in tests when constructing a portfolio. I can't remember this topic being discussed on this forum, so I thought I would share this with other members.
On another forum Palmer Wright wrote:Leslie Walko points to the potential danger of curve fitting caused by compounding. I agree, and have been concerned for years about how one market in a portfolio (commodity X) by being dramatically profitable in a single year can misleadingly bias the results of the whole portfolio.

During a multi-year test in TR, starting equity is low, perhaps $100,000, but compounding raises equity to many million in later years. The one-year outperformance of commodity X cand produce two kinds of curve-fitting bias: early-years bias and end-years bias. Mark Johnson's message describes the first, where X gives "a big turbocharged boost" to the portfolio's equity, which then gives a head-start boost to the number of trades in all the commodities traded. The second occurs when X's monster trades occur in the final years of the simulated time period when the large number of contracts makes X's profit far larger than if its big year came early. Here the profits contributed by X dwarf what they were in the first case.

As the message from M points out, we can avoid such biases by normalizing with a fixed-dollar bet size in testing to remove the galloping equity effect. I proposed this method in 1999, and still use it to compare with the compounded performance. I confess, however, that my testing has failed to find as much performance bias as I suspected I would find. The method is most important when selecting markets for a portfolio.

Palmer Wright
You can find the entire message here:

http://www.purebytes.com/archives/amibr ... 05313.html

Post Reply