Traders' Roundtable

I've added a countertrend system to a multi system trend following model and have gotten a 7% improvement in CAGR and a 5% reduction in drawdowns. It uses Stochastic crossovers to signal trades. Does anyone have countertrend systems that add value to trending systems?

I've tried, but never found any counter trend systems that add enough value to bother implementing.

LTTF systems suffer at the end of a trend. I found that adding a system that attempts to pick potential trend reversals (I used momentum divergences) enhanced returns EVEN THOUGH the top-picking system barely has a positive mathematical expectation.

This idea enhances overall return because, even though it doesn't work often, it works when you need it to - the LTTF system is about to give back a bunch of gains as the trend reverses but this is the time the trend-reversal system makes a profit. It straightens out your equity curve - Ralph Vince explores this issue as it increases your optimal f and thus increases CAGR.

I've had better luck with scaling out.

Thank you!

I employ 3 counter trend models in my strategy. 2 have contributed positively this year, while the 3rd was a loser. All 3 models have poor risk/reward metrics, yet when combined together, are acceptable. Even better, when they are combined with my other models, returns are enhanced (slightly), and volatility is dampened.

Do good counter trend systems have a relatively high accuracy rate?

The accuracy rates seem to be slightly lower and they seem poor as a standalone system. The risk:reward equations seem very poor compared to trend following systems.

What I cant figure out is how you short the tops, without also repeatedly shorting on the way up.

Question from Rob@everTrend: "Do good counter trend systems have a relatively high accuracy rate?"

Hard to give a general answer.
You will not find published counter trend systems.
All info on subject is anecdotal.

From the what its worth department: I know a NYMEX market maker with a several years of track record. He sold a counter trend system to Merrill-Lynch for few million. Accuracy was near 40%. Got a few million for it although SR was lousy. Had to disclose source and signed non-compete.

His current system's accuracy is a bit difficult to quantify since since he pyramids. Winning trades have very high profit factor. This guy firmly believes in multiplying number of entries by profit factor. His magic number is around 3. His current model is over 3. His annual rate of return ~20% with 15 % Max DD --so far.

Traded a system of my own, counter trend.
Out of sample accuracy was 72%, average of 3 years out of sample.
My profit factor is quite low compared to trend followers.
I don't get ten baggers.
Ten baggers are not the goal.

Completely different system pulled 60% accuracy out of sample, 3 years out of sample. I did not trade it--I should have.
I did not trade it since someone very very famous severely chastised my IN SAMPLE results. (However, they did offer to buy it for 'study purposes.'
Did not sell. My 3 years is up. I am recoding in TBB.

L

Longmemory

Why do you note the percent wins rather than returns and drawdowns for your systems? Percent wins is not very meaningful. Put a 0.5% profit target on anything and you will get about 80-90% wins but not high returns.

Rod

Hi Rod,

"Why do you note the percent wins rather than returns and drawdowns for your systems? Percent wins is not very meaningful. Put a 0.5% profit target on anything and you will get about 80-90% wins but not high returns. "

Percent wins is and easy to understand measure, but more to it. I evaluate all systems using several Reward to Risk Ratio method, e.g. quasi-Monte Carlo simulated equity stream divided by Nth percentile Draw Down, or actual trade outcomes (pessimistically adjusted for observational error) to the integral of the draw down surface, etc.

Reward to Risk Ratio measurements provide insight into "is System A better for me to trade then System B". They do not tell me anything about start of trade draw down and provide only hints bout "how likely are persistent runs against me".

Where probability of Win is P the likelihood of loss is 1 - P. If I begin to trade tomorrow, and my system wins 50% of time, it would seem that the likelihood of three consecutive losses is 0.5*0.5*0.5; likewise, chance of ten consecutive losses is 0.5^10. Seems like a small number, right? It is not since trading system returns are NOT random independent event but do show strong periods of runs. (You can read post like "we all had a very good year" or "last six months been difficult", etc.)

Frequent trading and probability winning systems are easy to track and operate: they can be monitored effectively since ten or twenty losing trades has minuscule probability of occurring. Keeps me from getting ruined while I await becoming rich.

Since I trade *frequent* short term traders, thousands of trades, this is a robust detection method.

L
p.s. Theory of runs, if discussed at all, is given childish treatment in finance textbooks. More sophisticated view is offered by Dietrich DORNER: The logic of Failure. For critique of finance 'research', see discussion of 'degenerating research' in Imre LAKATOS: "Falsification and Methodology of Scientific Research Programmes." in Lakatos + Musgrave: Criticism & The Growth of Knowledge.

Dorner provides cognitive model of what system failures are; Lakatos provides epistemology of how to avoid failure.

It also depends on what you define as countertrend. For example, I have found you can add to your expectancy by buying dips during up trends with quick exits. Some would call this buying against the immediate momentum countertrend, others would still call it trend following.

I also found Toby Crabels comments interesting when he said that he found high correlation always seemed to accompany similar time frames….He was more concerned with dividing things between long term and short term verses trend or countertrend.

Hi Dean,

"It also depends on what you define as countertrend. For example, I have found you can add to your expectancy by buying dips during up trends with quick exits. Some would call this buying against the immediate momentum countertrend, others would still call it trend following..."

Yes, "others would still call it trend following" makes sense since the set up involves some method of trend determination.

Trend following may be defined as 'expecting direction and skew of price data to be similar to post entry which can be observed pre-entry'.

Counter trend may be defined as 'expecting either direction, skew, kurtosis, stationarity, autocorrelation, heteroscedacity, power spectral density, or some other measure to change SUBSEQUENT to entry'. So 'counter trend' strategies may look as if selling tops and buying bottoms except the strategy does NOT have a concept of top or bottom.

L