System failure

M

minmike

I have been backtesting a pairs/regression to the mean, end of day strategy. I started to the backtesting in 1997. The system does extremely well until around 2003. Any ideas what could have changed? (It went from very profitable to almost random returns in 1 to 2 quarters time after having worked for 6~ years on paper.)

The profit going down so quickly seems to indicate to me that something happened other than people finding the system, but I could be mistaken. The system traded stocks.
 
my first thought, and this is something i noticed in the e-mini S&P is volatility. The average range of the day almost cut in half. Maybe the same for you?
 
Thats a good point and might be the case. The profit went from around $120 a trade to around $10 a trade. It also collapsed very quickly. (If it was just volatility contracting, I would have assumed that it would decreased slowly.)

It would be a profitable system at around $50 a trade.

Is there a good way to check? Maybe compare returns to the VIX? Its really frusterating because I broke the data into 3 sections for out sample testing, the first two were perfect, the last one was just horrible.
 
i can imagine how you feel, but around 2003 to 2004 the e-mini s&p range shrank very quickly. You may just want to put up a ATR on some of the stocks on a daily chart and look at their numbers.
 
what universe of stocks are you using to choose the pairs from? Is it possible that the good results in earlier years are unrealistic b/c your system picks pairs of penny stocks that could never have been traded profitably b/c of low volume/large spreads? I once had such a problem myself.
 
Markets change. It's how the business goes. When your system stops becoming profitable it's not because others are using it at the same time - others using your system would just mean more people trying to buy at the moment you're buying so you'd be more likely to not get filled as well. Eventually if a lot of people used that system getting filled with a market order would be so frustrating for everyone that only the person with the fastest connection would be able to use it profitably - but that person would still be able to use it profitably.

Systems deteroriate for fundamental changes you can't fight. In all the time you'll waste trying to figure out why what could have been never was, you can just develop a new system. Don't be too attached to the idea, it's just how the game goes.
 
No I didn't use penny stocks. I fact I drop any stock from the system if it drops below $20

Aaron
I will look into the ATR. That sounds like a good place to start. Any ideas on a way to salvage a system if that indeed is the problem?
 
Quote from minmike:

I have been backtesting a pairs/regression to the mean, end of day strategy. I started to the backtesting in 1997. The system does extremely well until around 2003. Any ideas what could have changed? (It went from very profitable to almost random returns in 1 to 2 quarters time after having worked for 6~ years on paper.)

The profit going down so quickly seems to indicate to me that something happened other than people finding the system, but I could be mistaken. The system traded stocks.
More...

How many trades were generated until the equity curve started going south?

Ron
 
You must log in or register to reply here.
Back
Top