Successful Trading with Algos Part 4

OK, you have created an algo and have picked the type of market you want to run it on with filters. It looks great over the last six months. I say six months because for a day trading algo you are not interested in how the market behaved two years ago as it's different now.


Here's an algo's results that I use on Light Crude.

Your next step is how to keep the algo in tune with the market. That means periodic reoptimization in a way that does not curve fit. I do that by testing a reoptimization regimen that walks forward. For example, I may take 20 trading days of data and reoptimize and then look at how it would have performed on the next 5 days of unseen data. I start this at the beginning of the six month period of back history and walk it forward every 5 days using the last 20 days of history and seeing the results on the next 5 days of unseen data. What I produce in MultiCharts looks like this:

 It's a pretty wide pic but let me draw your attention to some facts.

1. The original Nett Profit in the first pic is $13,008
2. When I add up the Out of sample (OOS) column profits the total is $6442 which is a more realistic of what would have happened using this algo to trade after reoptimising every 5 trading days using 18 days of historical data. The OOS P&L is the results of the algo on new data after having reoptimised on the 18 days before.

Doing the original work to get the results in #1 is just a first step. Getting a stable result from Walk Forward Analysis is what you need to get a more realistic representation of what is probable. Note that I say "probable", not possible because it's more certain than that but not "certain" because I cannot guess the future. What I do is trade with the odds - probabilities - very much on my side. 


Today's trade using that algo is below. The interesting thing is that you can run this algo in a "turn on and forget" mode or trade with it using it for entries and then manually manage the exits. Also, I have some different exit strategies that I can bolt on to manage the trades automatically to further improve profitability.

What I try and teach my students is to go through a process of building a potentially successful algo based on realistic expectations.