Stretch

This topic is empty.

Viewing 15 posts - 1 through 15 (of 37 total)

1 2 3 →

Author

Posts
August 24, 2020 at 10:22 pm #102048

TerryDunne
Participant

I’ve been thinking about this idea comparing MOC systems to a casino, that there is a small plus to be had so we want to do as many trades as possible to maximise those pluses.

The conclusions I have drawn from all of my back testing is 100% the opposite – we are not the casino but the card counting blackjack player, hopefully without being taken out the back and ‘educated’ by the more muscular employees of the establishment.

So, whenever there is mention of Nick’s “…the next 1,000 trades…” I took it to mean that we shouldn’t get hung up on one trade or a small series of trades because we have an edge and that will play out over the next 1,000 trades. However, I’ve come to wonder whether I’ve completely misunderstood…

Anyway, I decided to create a generic MOC system traded on the Russell 1000 – RSI(3) below 20, ranked by RSI(3). I turned off all filters, to simply see what happens when the stretch changes. These are some of the metrics when back tested from 1/1/2010 to 1/1/2020

Stretch Trades CAGR Drawdown
0.25 9,030 -5.45% 67.99%
0.5 5,404 9.77% 39.66%
0.75 3,137 17.12% 21.49%
1.00 1,741 10.42% 17.97%

So, my first question is have I done something wrong in my back testing so that these metrics aren’t reliable/accurate? This really worries me because all of my back testing shows the same general conclusion – the stretch is the edge – so if I’ve screwed something up all of my work over the last year is down the gurgler…

My next thought is, assuming the metrics are reliable, why is there such a focus on doing lots of trades rather than trading when the circumstances are most favourable?

I should mention, I’m not trying to have a go at anyone’s approach, just trying to get better at this.

Best wishes,

Terry

August 25, 2020 at 5:51 pm #112008

SaidBitar
Member

I agree with you
The stretch and ranking is everything at least this is what I noticed

August 25, 2020 at 6:50 pm #112009

Anonymous
Inactive

Hey Terry. I would say that the system using RSI(3) is about akin to me going into casino and playing blackjack – and I’m shit at blackjack.

Don’t assume that the RSI(3) is worth anything nowadays. There’s a reason why you can’t search up “best trading system” and get anything good… the systems you’d be seeking are not public knowledge. Shape up the entry criteria, and then I’ll agree with Said – stretch and ranking is extremely important, but only after the base has been built.

I wouldn’t turn off all filters. Nick justifiably railed me for thinking there was nothing wrong with buying 9999999 shares of some penny stock when I was attempting to design my first MOC system. You can’t do it live, so why act like it is an important data point in the backtest?

Hope it helps! Check my and Omar’s threads for some more info on MOC stuff. And Matthew’s, but best to have a coffee to go through that one. And your reading glasses.

August 25, 2020 at 9:06 pm #112010

ScottMcNab
Participant

Hi Terry,

My limited experience has also been that the stretch and the rank are critical components in these type of mean reversion systems…the further the prices has moved from the mean in the short term then the higher the probability it will revert. From here, however, the design of systems seem to vary widely as seen by the variation of returns members obtain in different months….when some do great, others lose and then this is reversed the next month.

My systems also tend to lean towards doing fewer trades and looking for specific situations when the mean reversion is more likely to occur. My mean reversion systems have only done <300 trades combined for the entire year but have a profit factor of 1.7 and win/loss of 1.1. If I was able to come up with a different type of system that did lots of trades I would add it for diversity.

IMHO if your system has passed all the tests then feel good about it. Over the years, as you come up with different ideas, add them to the mix. I still have RSI as a condition in some of my systems but don’t tell anyone.

August 25, 2020 at 11:56 pm #112015

TerryDunne
Participant

Hi Seth,

I think I’ve miscommunicated. I’m not suggesting that this is even a system, let alone a good one.

My aim was only to compare different stretches. I expect that a proper system would perhaps reduce the difference between the different stretches, but my issue was more about whether a ‘small stretch’ to do more trades or a ‘big’ stretch to do better trades was the best strategy.

Best wishes,

Terry

August 26, 2020 at 12:20 am #112018

TerryDunne
Participant

Hi Scott,

Your point about different members having wide variation of returns is an important one I suspect. This implies that the Law of Large Numbers applies, so doing as many trades as possible will smooth out this volatility.

However,the more trades done, the less selective one can be and the more damage is done to the ‘edge’…

I have similar numbers to you – only properly live this month, but so far actuals are replicating the matching back test pretty closely, so I’m cautiously optimistic that my back test results are reliable.

I’ve also found that only trading on days where the index is ‘suitable’ really enhances results.

All the best,

Terry

August 26, 2020 at 12:22 am #112014

TerryDunne
Participant

Hi Said,

I’m getting to the point where I’m thinking about a system with no entry criteria at all, just a ranking and a stretch…I’ll let you know how it goes.

Regards,

Terry

August 26, 2020 at 2:27 am #112020

JulianCohen
Participant

I have a number of MOC systems, most of them the smaller the stretch, the larger the CAGR and greater the MDD, so I tend to choose an area, normally around 0.3 or 0.4 that gives me a decent balance.

The others definitely have an optimum stretch, normally between 0.4 and 0.8, with performance dropping off before and after those areas.

I’ve noticed that my systems that look for an expansion in volatility need a smaller stretch but for the ones that use other criteria for entry this doesn’t necessarily apply. I also keep my systems as simple as possible, with one or two main parameters.

YMMV of course, but that’s what I’ve noticed with my systems.

August 26, 2020 at 4:06 am #112022

Anonymous
Inactive

I like the large volume of trades versions of these systems for a few reasons.

Main ones being more positions means less risk per position so if any one thing goes wrong with earnings or announcement gaps I don’t panic too much. Luck is a factor in any method of trading. Having only a fraction of your account risked per position means you could probably do 50 losing trades in a row and it still wont catastrophically damage your account too much. Greater volume means reduced chance of being trapped by the unlucky moves.

Also because of human impatience combined with expectancy of large numbers. If you can only expect to see statistics/expectancy of any type of accuracy play out only after 1000 trades or more, if you are only doing 300 trades a year in total it is going to take you three years to work out if you are satisfied that the system is moving in the right direction for you. Whereas doing 3000 trades a year means after four months you’ll know if you’re going in the right direction or not. I don’t have the patience to see if trading systems are working over three years and then adjust. I’ll give something six months tops before I start to get impatient and seek out improvements.

Earlier versions of my systems had lower trade frequency and there were some days where I wouldn’t get any trades whatsoever. I think there might have even been a week or two in there with no trades. To work on a system each day and not see any trades can be a mental challenge. I even started to think sometimes “Is it broken and not working properly?” when of course it wasn’t. It may seem juvenile or prove my inexperience, but seeing trades each day having been made simply makes me think I’m actually doing something and working for the trades and the work I do each day is actually creating an effect.

Finally my biggest obstacle of lower volume systems in the past was that it simply meant bigger positions required. Even in the most liquid market in the world, I was always finding that trying to get fills on large volume orders at my target price was difficult. I rarely try to order any more than about 2000 units per position now to help combat this so I can have a higher chance of getting my fill at the target price.

I see this type of system as me being the casino with as many roulette tables jammed into my room as I possibly can (my trading account) and the symbols I trade are the punters coming to my tables and having a go at the odds of winning. And instead of having a wheel with 18 red, 18 black and one green zero with 1:1 odds on the red/black bet I’ve got more like 17 red, 17 green and 3 green zeros with only 1.0:0.9 odds on the red/black bet. I’d be pretty sure most humans wouldn’t gamble on tables with odds like this but if the Russell 3000 constituents are prepared to turn up at my establishment day in and day out and play these tables of mine with these odds loaded in my favour then I’ll happily take their money over a large volume of trades.

August 26, 2020 at 8:22 pm #112024

ScottMcNab
Participant

Matthew O’Keefe wrote:

Also because of human impatience combined with expectancy of large numbers. If you can only expect to see statistics/expectancy of any type of accuracy play out only after 1000 trades or more, if you are only doing 300 trades a year in total it is going to take you three years to work out if you are satisfied that the system is moving in the right direction for you.

Hi Matthew. I also like 1000 trades, if possible, to have necessary confidence but for me this is in the back test …once live, I just want to see the live results agree with the back test and for the metrics to remain similar (ie the fluctuations from year to year remain within extremes seen in back test).

August 26, 2020 at 11:13 pm #112023

TerryDunne
Participant

Hi Julian,

Interesting insight re range expansion triggers, I’ll have to investigate further. I actually have a ‘rule’ with stretch, I will only use ‘major’ fractions (1/4; 1/3; 1/2; 2/3; 3/4; 1), paranoia about curve fitting probably!

I have no idea what ‘YMMV’ is – might need to ask my teenage children

August 26, 2020 at 11:30 pm #112025

TerryDunne
Participant

Hi Matthew,

I think your post really highlights how important it is to match systems to individual preferences. For you it seems to be action, for me it’s fear of loss. My disastrous experiences during the GFC mean I’m (not quite) comfortable sitting on the sidelines, but happy to know that my back testing shows draw down no worse than in the mid teens. Almost coincidentally, this seems to maximise return as well, at least for my systems.

Best wishes,

Terry

Regards,

Terry

August 27, 2020 at 4:18 am #112026

JulianCohen
Participant

Your Milage Might Vary

I use stretch in 0.1 increments. Seemed easy when I started.

August 27, 2020 at 10:55 am #112027

TerryDunne
Participant

August 27, 2020 at 4:03 pm #112019

Anonymous
Inactive

Terry Dunne wrote:

Hi Seth,

I think I’ve miscommunicated. I’m not suggesting that this is even a system, let alone a good one.

My aim was only to compare different stretches. I expect that a proper system would perhaps reduce the difference between the different stretches, but my issue was more about whether a ‘small stretch’ to do more trades or a ‘big’ stretch to do better trades was the best strategy.

Best wishes,

Terry

Ah – I’ve got you now. So what I would say regarding how tight your stretch parameter is: this depends dearly on your ranking [and maybe main parameters]. A good example is to take your current ranking (or maybe one that ranks by positive ROC) that uses a .3 or .4 multiplier and swap out the ranking for ATR/C. You’ll see you need a VERY LARGE multiplier for a system like that, so the stretch number varies largely based on the internals of the system.
Author

Posts