You are using an out of date browser. It may not display this or other websites correctly.
You should upgrade or use an alternative browser.
You should upgrade or use an alternative browser.
When to turn off and turn back on a strategy?
- Thread starter mizhael
- Start date
Edit the source code of each strategy and introduce an externally supplied input value, named Doomsday. If Doomsday is equal to 1, exit all positions immediately and do not enter any new positions. If Doomsday is not equal to 1, operate normally.Quote from mizhael:
what's a good way to turn off and then turn it back on a strategy?
To turn off a strategy, merely change its value of Doomsday to 1.
To turn it back on, change its value of Doomsday to zero.
When a strategy is underperforming expectations, turn it off. When it's performing as expected, turn it back on.
I run multiple strategies in real time simulated. These strategies take more trades then my real trading. When my performance is less than I would like, I check my simulations to see if the problem is with my specific trades or the strategies. In Jan/Feb, my longs and shorts were underperforming. My long sims were also underperforming but within expectations, so I rode it out. On the short side my real trades were underperforming but the sims were not, so I made some adjustments in my trading.
Back in the summer of 2008 almost all my longer term strategies started breaking down, so I got out and focused on what worked, which was my shorter term strategies.
I run multiple strategies in real time simulated. These strategies take more trades then my real trading. When my performance is less than I would like, I check my simulations to see if the problem is with my specific trades or the strategies. In Jan/Feb, my longs and shorts were underperforming. My long sims were also underperforming but within expectations, so I rode it out. On the short side my real trades were underperforming but the sims were not, so I made some adjustments in my trading.
Back in the summer of 2008 almost all my longer term strategies started breaking down, so I got out and focused on what worked, which was my shorter term strategies.
I have been experimenting with the following (aka "the JoeKrut Diff" http://etrackrecords.com/index.html):Quote from mizhael:
... what's a good way to turn off and then turn it back on a strategy?
1. Track nominal account equity for each strategy on an ongoing basis (for 1 contract, or 100 shares, etc).
2. Calculate a 30-day simple moving average of this nominal account equity (assuming it's a strategy that trades 2 or 3 times each day ... experiment with different periods for different trading frequencies).
3. If nominal account equity drops below SMA(30), stop trading the strategy live, but continue trading it in simulation.
4. Continue to keep track of what nominal account equity would have been if you had kept trading it live.
5. Re-instate the strategy live once nominal account equity increases above SMA(30).
Supposedly, this system will keep you trading the strategy when the strategy is "in tune" with the market, and stop you when it is not.
However, I recently had an unpleasant experience with (until then) a very profitable strategy that suddenly started losing consistently. When the losses came, the above strategy management methodology stopped me trading the strategy live. I monitored it in simulation until I got the signal to put it live again (and I saw what I thought was the good ol' successful performance coming back). I then put the system back live, and â guess what? - immediately it performed badly again. So I put it again back into simulation. And so on ... for two more whole cycles of the same!!! The system management methodology was âwhipsawingâ me, i.e. keeping me out of winning periods, and only putting me back in time for the losses! LOL!
After this, I changed tack. The truth was, initially I had stumbled across the strategy by luck, without really understanding why it was (initially) so successful. Trying to figure out why it had stopped working so suddenly, I read some stuff that helped me understand better why the strategy might have worked so well in the first place, and this suggested adding a new condition for my signal... which I did. And since this time, the strategy works great again. (Was I "over fitting"? Not in my opinion, because there is a consistent "Cartesian"/"Newtonian"/deterministic/logical reason for adding the new condition).
So, what moral did I drawn from this?
When a good strategy stops working, IMHO it is best to try to figure out why this might be. If you can get to understand why, you may be able to change your signal logic sensibly in such a way that the strategy performs more robustly going forward.
The drawback of a "JoeKrut diff" type methodology IM again HO is that it's a shortcut alternative to putting in the time to understand what the underlying problem is. Also - being a sort of lagging indicator - it lets you take a string of losses before signalling that you should stop trading the system, and then lets you miss a string of winners before putting the system back in play...
Quote from MGJ:
Edit the source code of each strategy and introduce an externally supplied input value, named Doomsday. If Doomsday is equal to 1, exit all positions immediately and do not enter any new positions. If Doomsday is not equal to 1, operate normally.
To turn off a strategy, merely change its value of Doomsday to 1.
To turn it back on, change its value of Doomsday to zero.
How do you decide this external variable? Based on your own view about the market?
It's this question I am asking...
How do you judge when to stop a strategy and when to turn it back on.
And when to change/refine a strategy?
A strategy that in theory can work well long term may very possibly perform poorly in short term.
During that very short term, we should do something about it, right?
We cannot afford just sitting there and do nothing and see our capital/wealth deminishing, right?
Who knows if the strategy will come back again, or it is just permanently bad (even though in backtest, it looks good with the possibility of overfitting...)
Quote from dwpeters:
When a strategy is underperforming expectations, turn it off. When it's performing as expected, turn it back on.
I run multiple strategies in real time simulated. These strategies take more trades then my real trading. When my performance is less than I would like, I check my simulations to see if the problem is with my specific trades or the strategies. In Jan/Feb, my longs and shorts were underperforming. My long sims were also underperforming but within expectations, so I rode it out. On the short side my real trades were underperforming but the sims were not, so I made some adjustments in my trading.
Back in the summer of 2008 almost all my longer term strategies started breaking down, so I got out and focused on what worked, which was my shorter term strategies.
How do you compare the realized performance of the strategy to your expectation, given the future may not be similar to the past at all?
How do you form your expectation in the first place?
And once you decide a strategy is not working, you turn it off, and then when do you turn it back on? Based on your view again?
Quote from mizhael:
Lets say you have a bunch of strategies that are currently running in real-trading,
what's a good way to turn off and then turn it back on a strategy?
I'm reminded of a quote from Ed Seykota's Market Wizards chapter. It goes something like:
If you have a successful method, M, for deciding when to modify your trading system S, you should consider just trading M.
I probably butchered the wording of that, but his point stands.
Quote from abattia:
I have been experimenting with the following (aka "the JoeKrut Diff" http://etrackrecords.com/index.html):
1. Track nominal account equity for each strategy on an ongoing basis (for 1 contract, or 100 shares, etc).
2. Calculate a 30-day simple moving average of this nominal account equity (assuming it's a strategy that trades 2 or 3 times each day ... experiment with different periods for different trading frequencies).
3. If nominal account equity drops below SMA(30), stop trading the strategy live, but continue trading it in simulation.
4. Continue to keep track of what nominal account equity would have been if you had kept trading it live.
5. Re-instate the strategy live once nominal account equity increases above SMA(30).
Supposedly, this system will keep you trading the strategy when the strategy is "in tune" with the market, and stop you when it is not.
However, I recently had an unpleasant experience with (until then) a very profitable strategy that suddenly started losing consistently. When the losses came, the above strategy management methodology stopped me trading the strategy live. I monitored it in simulation until I got the signal to put it live again (and I saw what I thought was the good ol' successful performance coming back). I then put the system back live, and â guess what? - immediately it performed badly again. So I put it again back into simulation. And so on ... for two more whole cycles of the same!!! The system management methodology was âwhipsawingâ me, i.e. keeping me out of winning periods, and only putting me back in time for the losses! LOL!
After this, I changed tack. The truth was, initially I had stumbled across the strategy by luck, without really understanding why it was (initially) so successful. Trying to figure out why it had stopped working so suddenly, I read some stuff that helped me understand better why the strategy might have worked so well in the first place, and this suggested adding a new condition for my signal... which I did. And since this time, the strategy works great again. (Was I "over fitting"? Not in my opinion, because there is a consistent "Cartesian"/"Newtonian"/deterministic/logical reason for adding the new condition).
So, what moral did I drawn from this?
When a good strategy stops working, IMHO it is best to try to figure out why this might be. If you can get to understand why, you may be able to change your signal logic sensibly in such a way that the strategy performs more robustly going forward.
The drawback of a "JoeKrut diff" type methodology IM again HO is that it's a shortcut alternative to putting in the time to understand what the underlying problem is. Also - being a sort of lagging indicator - it lets you take a string of losses before signalling that you should stop trading the system, and then lets you miss a string of winners before putting the system back in play...
Is this Joe Krut training classes good? Anybody had experiences?
I also had the same experience with using moving average to turn on and off strategies, yeah, it's problematic. That's why I am seeking advice.
First of all, how do you judge if the realized performance was due to (1) tmporarily changing market conditions; or (2) the strategy is fundamentally flawed?
Moreover, per your second point in your posting, if my one strategy has many instruments bundled, how and what's the best approach to dig into why it would not work?
For example, the big problem I have is:
Lets take momentum strategy, when does it work, and when does it not?
Under what market conditions?
If we can find it out, then we can turn on and turn off it thru a scientific approach...