Welcome to NexusFi: the best trading community on the planet, with over 150,000 members Sign Up Now for Free
Genuine reviews from real traders, not fake reviews from stealth vendors
Quality education from leading professional traders
We are a friendly, helpful, and positive community
We do not tolerate rude behavior, trolling, or vendors advertising in posts
We are here to help, just let us know what you need
You'll need to register in order to view the content of the threads and start contributing to our community. It's free for basic access, or support us by becoming an Elite Member -- see if you qualify for a discount below.
-- Big Mike, Site Administrator
(If you already have an account, login at the top of the page)
I've shared a strategy I've been working on with Zoethecus, and this morning he asked if I thought it's lost its edge, because it's establishing a new draw-down level. I didn't answer right away because I've been thinking about it all day.
I decided to post my answer publicly because I think it merits discussion (I think I could learn from other's opinions on the matter).
As this strategy is still experimental, I'm not even sure yet that it has a generalized edge. I'm qualifying with 'generalized' because that is what I seek in a strategy. By this I mean it doesn't just have an edge in rising markets, or just in falling markets, or just in trending markets, or just in chopping markets, etc. I want it to have stable performance over a broad range of market conditions.
Once a strategy is found which has a generalized edge, I don't think it's likely to ever loose its edge until such a time as it becomes widely followed, or market structure changes.
By market structure I mean something like the introduction of a new instrument which changes long established arbitrage relationships, or something of that sort.
In seeking a generalized edge I look for a balance between short and long performance. I think it would also be a very good idea (but I don't have the tools or patience for it) to look for low correlation between performance and volatility, unless the strategy factors in volatility.
I score my strategy backtests with a personal variation of SQN.
Getting back to the original question, I don't think I would abandon a strategy just because it is setting new draw-down levels, unless the new maximum draw-down resulted in a new SQN score which I found unacceptable. In that case I'd drop it in a heartbeat.
As long as the new SQN score is still attractive, it is my nature to be persistent to a fault, and I would continue trading it with the expectation that its performance should return to the the established expectations.
There's a saying among many strat traders that goes along these lines: by the time you realize a strat has lost some or all of its edge, you've experienced a drawdown way beyond what you envisioned.
Others, who have witnessed long periods of smooth equity curves, will cut the strat loose when it breaks a major trendline. (I don't not subscribe to this maligned application of TA.)
If you're so inclined, can we learn more about your SQN?
I'm still trying to find the ideal fitness function. Net Profit / Drawdown as suggested here previously doesn't do it for me, as that ignores consistency and unrealized risk. The attached is the best I've come up with so far. For me the goal …
Unfortunately for every system that you have backtested its biggest drawdown is yet to come. ....that does not mean the edge has gone, it might just be on holiday.
While I am not a massively systemised trader or backtester, I understand a lot of the issues involved, and as a result I have never been a purist believing that one size fits all.
It seems that the old area of diversification of systems is what is often required to smooth the overall portfolio PL of all the systems. Alternatively use a discretionary element (which is off course not able to be backtested) in order to determine the context of which systems to apply based on the market or instrument.
This idea of applying context, and the psychological aspects of being able to ride out drawdowns is an often overlooked element that is part of the edge of a system.
(as an extreme example - remember when Warren Buffet was told he was a dinosaur and that he had lost his edge and was not with the new economy about ten years ago......well last I heard he was still 2nd richest guy in the room)
For what it is worth, I follow Pardo's approach in continuous assessment of real time performance (not limited to DD, but including %win and W/L ratio) versus the 'idealized' model performance; in other words, is the live performance within a predefined range of acceptability versus the historical performance of the model?
How to establish a range of acceptability is discretionary and depends on your appetite for risk. The range of acceptability is essentially a hard limit or cutoff point beyond which your system will be deemed to be dysfunctional and should be discontinued. A statistical approach is easiest means of establishing range limits; one such approach is based on the 68-95-99.7 rule.
I personally use a 2 SD limit (cutoff): About 95% of the values lie within 2 standard deviations of the mean (or between the mean minus 2 times the standard deviation, and the mean plus 2 times the standard deviation). Therefore, if your parameter (current DD, for example), falls outside the limit of the mean plus or minus 2SD, this represents a substantive departure from the norm (because it should occur less that 5% of the time) and COULD represents a substantive departure from the expected performance. In my case, I would shut down the strategy if this occurred. Alternatively, you could be more conservative and use a 1SD limit, or less conservative and use a 3SD limit.
As an example of how I use this approach, here is a chart in which the live trading W/L ratio is plotted to assess where it is tracking relative to the model W/L ratio. The model value is based on the performance of the same system backtested over the exact same trading period as was traded live, i.e. this is a comparison of real life (live) versus idealized (backtested model) performance. As long as the live W/L ratio is within the 2SD bands, I am confident that the live W/L ratio is within the range expected based on the historical performance of the model. As you can see in the chart, the live W/L ratio is very close to the mean value produced by the model.
Notice how there is initially a high degree of volatility in the live W/L ratio, which diminishes as the total number of trades increases. This is one reason why a lot of trades are needed before even attempting to use a statistical approach.
Great post, really gives me something to think about. One thing I'm unsure of is how to apply it to drawdown. Would you apply the stddev of the per trade returns to the maximum drawdown, or would you only use loosing trades?
