Wednesday, January 10, 2007

2 Elements of Good Backtesting Results

We're not the only ones on the subject of backtesting in the new year. Today Price Headley points to the benefits on a trader's psychology and confidence of backtesting and helps frame how a trader should evaluate the system to which s/he is about to commit capital (thanks TraderMike):
  • How will your system react to drawdowns?
  • Is your trading system practical? (i.e. do you have time to use this system?)
  • Can you tolerate inactivity if your system suggests doing so?
Source: "The Truth About Testing Trading Systems", Price Headley

Yesterday we outlined steps for taking backtesting lessons into the real trading world using The Odds Maker as an example (the steps, in general, can be applied to other tools). Today we focus on two necessities of a good backtesting result: the win rate and the presence of a large spread between average winners and losers.

Here's an apt analogy about backtesting any system:
Location, location, location : Real Estate :: Spread, spread, spread (between the system's average winner and average loser) : Relevant Backtesting (in the real trading world)
This applies to any backtesting method - whether you are evaluating how an individual stock performed or a entire strategy performed.

Keeping the spread wide enough is a cushion against the other trading parameters that over complicate most systems (e.g., slippage, commissions, etc.). These are important elements to the success of a system, but avoid taking a micronmeter to a mudpuddle and focus more on finding a spread (or ratio) good enough to account for them.

Elements of good backtesting results:
  1. A success rate of over 55% in either direction (i.e., Long/Short)
  2. A significantly large average winner to average loser ratio (i.e., a ratio of 1.5 or more where, for example, winners average $1.50 and losers average $1.00)
UPDATE: This isn't THE benchmark to grade all other systems - just an indicator that you might have something that works. You could have a win rate much lower but compensated by an even larger spread between average winner and average loser.

For an excellent example of this read John Forman's response to this post. His point is that there is a bigger picture to consider and that the suggested values above are not to fixed at a set point. Rather they relate to each other in different ways. More about this in our next post.

Here's a clear picture of what mean in a strategy summarized by The Odds Maker:




The Odds Maker:

The Odds Maker finds tradable patterns (built using Trade-Ideas alerts and filters) with high odds of success and summarizes rules on how to take, hold, and exit a position. One has to follow the system to come close to the result summary at the bottom of the alert window. Ignore these position rules at your own capital’s risk. The more you deviate from doing exactly what the strategy suggests, the higher the chance your trading results will not match the results of the system.

2 comments:

Anonymous said...

I would also add that it is important to evaluate how realistic your backtesting is. For example have you taken into account adequate slippage and commissions? And most importantly if you've optimized your system (even manually) you need to test how it performs out of sample.

D TradeIdeas said...

Great question. I answer your first very valid point in the next blog post. With regard to testing systems out of sample, we do extensive testing that we will soon allow our subscribers to do as well.

The Odds Maker produces a list of the trades the system generates and then presents the summary analysis to the trader. In between these steps The Odds Maker divides the results into odd and even numbered results. Each list gets tested to ensure there are no biases. When we make this available to our subscribers it will be called a random filter.