I have been developing and testing several systems in the past few weeks with mixed results. The systems are performing well from 2007 - 2008.. in fact, they are performing extremely well. However, from 2006-2007 they go on losing streaks. I wanted to know if anyone else is having this issue with their back-tests. 2006 - 2007 is when the economy was going through an exceptionally turbulent economy (mortgage crisis, oil). To be fair, the economy still isn't good but at that time it was more of a surprise and reactions to the fundamentals were more exaggerated. I have several methods built into my systems to adjust entries and exits for volatility, but 2006-2007 is just really giving me a bashing. Despite this volatility adjusting, the systems are still losing from 2006-2007. I know that a lot of quantitative based funds took a hit during that year so I am fairly certain it is not just me. However, I am wondering how some of these quantitative systems have been adjusted to prepare for such an event at a future date.
How long is your typical holding time for a trade? Higher speed systems should inherently be more immune to changes in the general market IMO.
The majority of my systems - ranging from intraday high frequency to those with multi-day holds - are incurring losses during the 2006-2007 back-test period. From pattern-based systems to mean-reversion, nothing is immune. I am working on a way to better measure volatility allowing for the systems to more effectively adapt to market shifts, but until then I am seeing if anyone has any other solutions. It should be noted that the systems made MORE than enough during 2007-2008 to cover any losses incurred before that period. I generally don't risk more than 0.2% on any one trade so drawdown amounts aren't significant and I can tolerate such events. Still, a year long draw-down is no walk in the park.
Your systems are flawed. Don't rationalize or justify the fact that some of your logic is incorrect. Go back to the fundamental concept and keep reworking it. Down years for intraday systems are not acceptable. Saying something like "nothing is immune" is silly. A good intraday system is immune to general market conditions and should perform BETTER with higher vola.
I do agree with you. I do not want to justify a flawed logic base. Emotions such as pride and ignorance do not do well in the system development field. I'm taking the algorithms back to the drawing board to see if I can get them to intrinsically factor in volatility rather than passing the trades through a secondary volatility filter.
So I was manually looking through the charts and I noticed some irregularities. The continuous contract I was using to run the back-tests did not line up with the actual contract. Looking through, the continuous contract chart contained bars with 60 pt ranges from top to bottom where the regular contract had no such thing. This was happening repeatedly throughout the chart with enough frequency to mess up the systems logic. Turns out the historical data I was using was corrupt. The abnormal, flawed high range bars were throwing off the systems' algorithms causing them to improperly determine volatility levels, affecting entry and exit levels. After fixing this, the systems are profitable during the 2006-2007 year. Not as profitable as 2007-2008, but still within predicted levels. I generally update my historical data files every week, but it turns out that only the regular contracts were being updated and not the continuous contract. Moral of the story: Make sure you check your data and refresh your database often.
when does 2006-07 period end for you and when does 2007-08 start? around jul-07 VIX spike? could you check your system on 2000-02 period to see if it also outperforms?