The curse of systems trading/Why does this happen
- Thread starter Robertwiz
- Start date
Statistics don't support it. December is statistically a good month.
Buy in December and sell in April:
http://www.ritholtz.com/blog/2009/04/sell-in-may-and-go-away-fact-or-fallacy/
There is a very simple reason for this: you are a still in the mindset of a discretionary trader. Instead of trading a single stock you are trading you TF system's equity curve with the same (unhappy) results that prompted you to use it in the first place.
You either trust the back test or formulate money management rules that you (psychologically) can live with.
I'm with you - I don't want a drawdown greater than 30% but some of the largest gains seem to come after a huge drawdown.
Reducing the DD with equity curve stops takes some of the heat off you and allows greater leverage, if you're in the mood.
Solution: paper trade the sys when it's not in synch, give it the football when it's starting to run..........
If volatility is greater than a threshold amount, reverse your buy / sell signals.
http://www.ljmu.ac.uk/AFE/AFE_docs/ART0105.PDF
My guess is that you don't truly understand the source of your edge (if there is one), so you aren't catching some variable that really dictates the performance.
Please understand that I am not suggesting there's some flaw with you, I'm just saying there is a hidden variable you haven't isolated.
One of the common faults in developing automated systems is a trader does not understanding the characteristics of price data (metadata) they use to test their system with. One of the first trend following systems I wrote about 20 years ago taught me this lesson. The story goes like this:
Back in the early days of Omega Research before 2000i I wrote and back tested (with 7 to 10 years of data) a trend following system that produced great results. In the trends of the late 1990s the system worked to perfection just as testing said it would. However when congestion hit losses were 2 or 3 times the projected amount and whipsaws killed all of trend gains and then some.
In frustration I almost threw out this system thinking I had produced a real dud. But being an IT person I decided to do what we call a âlessons learnedâ session and try to find the mistakes I had made and log them. What I uncovered help me change this âlosingâ system into a winner that I still use today.
The problem turned out to be volatility. This means the volatility of the price data in back testing (mid 1980s to mid 1990s) was not the same as the volatility I experienced in live trading in the late 1990s. The increased volatility of the market in the late 1990s at time pushed trade gains and losses much higher than occurred in testing. And in this case the losses in congestion were way out of line and wiped out the gains of the trends.
I found I had two choices. Wait to trade the system in conditions similar to those of the original back tests or modify the system to handle different levels of volatility. I chose to modify it to handle volatility levels. This turned out to be much harder than I anticipated. I am still adding âdynamicsâ to these old systems.
My point is beginning system traders, like I was in those years; seem to think they can throw any price data you want at a system when testing it and it will live trade just fine. While this may be true for simple set up trading it does not hold up, as I can attest, with more complicated trading such as trends. I order to have faith in your system you have to know what kinds of markets and market conditions best suit your system.
Back in the early days of Omega Research before 2000i I wrote and back tested (with 7 to 10 years of data) a trend following system that produced great results. In the trends of the late 1990s the system worked to perfection just as testing said it would. However when congestion hit losses were 2 or 3 times the projected amount and whipsaws killed all of trend gains and then some.
In frustration I almost threw out this system thinking I had produced a real dud. But being an IT person I decided to do what we call a âlessons learnedâ session and try to find the mistakes I had made and log them. What I uncovered help me change this âlosingâ system into a winner that I still use today.
The problem turned out to be volatility. This means the volatility of the price data in back testing (mid 1980s to mid 1990s) was not the same as the volatility I experienced in live trading in the late 1990s. The increased volatility of the market in the late 1990s at time pushed trade gains and losses much higher than occurred in testing. And in this case the losses in congestion were way out of line and wiped out the gains of the trends.
I found I had two choices. Wait to trade the system in conditions similar to those of the original back tests or modify the system to handle different levels of volatility. I chose to modify it to handle volatility levels. This turned out to be much harder than I anticipated. I am still adding âdynamicsâ to these old systems.
My point is beginning system traders, like I was in those years; seem to think they can throw any price data you want at a system when testing it and it will live trade just fine. While this may be true for simple set up trading it does not hold up, as I can attest, with more complicated trading such as trends. I order to have faith in your system you have to know what kinds of markets and market conditions best suit your system.
no kidding, trading everyday is like sitting under a tree, it looks like the same tree but it's always growing. I took five years off from the market, and when I came back the change was very obvious, the old way still worked, but all the levels had to be adjusted to the changed market. The challenge is designing a system where absolutely nothing is static and everything is always changing.
