The Non-Optimisation Myth

[BlueHorseshoe 10]

Your trading system uses exponential moving averages, a volatility indicator, market profile, and an oscillator to identify profit-taking opportunities.

 

I scoff at you because I know that your strategy is simply the result of over-optimising the parameters of a load of indicators, and that though you may have curve fitted them perfectly to historical data, your results in live trading will not be good.

 

I'm much smarter than you - my system is based on price alone. I simply short the ES at the prior day's high, and buy the market at the prior day's low. I don't care about defining trends with a 76 period EMA, or using Average True Range to trail a stop-loss. No optimisation there, thank you very much!

 

Can you work out why I am a fool, and why I am deceiving myself?

 

I should be asking myself questions such as the following:

 

- Why yesterday's high? Why not yesterday's high plus a bit? Or close?

- Why the ES? Why not Gold ETFs? Or Lean Hog futures?

- Why not the low two days prior? Or five days? Or thirty?

- Why not buy at the high, and sell at the low? Why not be flat the entire time?

- Why a daily chart and not a five mintute chart? Or a monthly chart?

 

All of these are examples of parameters that I have selected (pressumably because I believe them to be optimal in some way) - If I think that I have escaped the possibility of curve-fitting then I am just deluding myself. Whenever a trader designs a system/strategy etc, they make decisions which amount to a form of optimisation. The dangers of curve-fitting can be intelligently minimised; they cannot be eliminated entirely.

 

BlueHorseshoe

[Predictor 11]

Blue, curve-fitting is actually a goal when designing trading systems -- i.e you want your methods to "fit" the data. What you don't want to do is to OVER FIT the data which makes it less likely to work in the future...

 

Optimization is a goal for most system developers. I know it is for me.

[BlueHorseshoe 10]

Blue, curve-fitting is actually a goal when designing trading systems -- i.e you want your methods to "fit" the data. What you don't want to do is to OVER FIT the data which makes it less likely to work in the future...

 

Optimization is a goal for most system developers. I know it is for me.

 

Hi Predictor,

 

I agree with you. There's a very distinct difference between optimisation and curve-fitting. There's a great quote from William Eckhardt on this topic that I'll have to try and dig out . . . My opening post, however, is directed for the benefit of anyone who thinks that optimisation can be circumvented altogether (I was once one of those people) - perhaps this would just be newer traders.

 

If a system is based on robust concepts and a sound description of market behaviour, it should prove profitable across a broad range of parameters.

 

A more interesting question to ask somone such as yourself might be: once you've optimised your strategy (without curve fitting it), how would you seek to improve its profitability without further optimisation? For me (having already explored money management approaches) the first answer is to look at order execution, but I'd be interested to hear anyone else's thoughts.

 

BlueHorseshoe

[Predictor 11]

Well, one could say order execution is another form of optimization. Certainly one is introducing new variables/things that can change/unknowns...

 

You may to look at my article on Good vs Great Trading. I feel that knowing how to push winning trades is probably the best way to enhance a system. Its not easy though. I primarily apply the principle using my discretion.

 

If one could find a variable that correlates roughly to the probability the trade will work out (or profitability of the trade) then that could be used as a confidence ranking and one could vary the position size based on the confidence (the probability the trade would work out)-- increasing size on better trades and decreasing it on the more average trades. I have been meaning to apply this concept in some of my programmed systems.

 

A variation of this concept could be used to trade a system through a variety of parameters but with different weightings to achieve a blended performance, i.e to try to emphasize the parameters that are believed to work best while taking advantage of robustness of additional params.

Edited August 27, 2012 by Predictor

[onesmith 10]

Whenever a trader designs a system/strategy etc, they make decisions which amount to a form of optimisation. The dangers of curve-fitting can be intelligently minimised; they cannot be eliminated entirely.

 

That's true for most developers because they fail to make the a priori choice to design an algorithm to guarantee specifically that it's "entirely eliminated".

[onesmith 10]

once you've optimised your strategy (without curve fitting it), how would you seek to improve its profitability without further optimisation?

 

Locate and optimize additional signals that are at least somewhat different than the primary signal.

[SIUYA 11]

Locate and optimize additional signals that are at least somewhat different than the primary signal.

 

diversify???........

 

Personally ( and these are thoughts and feelings rather than anything else......)

 

The problem with any model for how the market works is that it will always hit difficulties in certain types of markets. ie; no one thing works all the time, there is no truth to the markets, models are just idealised....and anytime people think the model represents the truth - Mr market might remind you otherwise.

 

so why not optimise and then keep optimising as the market changes - could you run a test that optimises within a range daily and adjust - what difference would that make?

My guess is that if the general theory of how the market works is correct, then all you are doing is optmising entries and exits for that market at that time (a 20 period MA as opposed to 25MA will sometimes make a difference, othertimes not) - you wont change how the market works....it will do that on its own as it always adapts.....so why not adapt with it. (the key is getting on and off at good times for the market state at that time)

 

Also as different instruments and different markets have different participants and underlying drivers, liquidity and order flows, why would we expect one model to be robust for all of them - one theory of market behaviour maybe - but not one model and ideal set of parameters.....as markets dont have the set rules of a casino, and they dont follow set distributions then why should we think they will?

 

Even risk managers fall into the same trap - VAR analysis - what a joke - always has been for years, and yet it was embraced by risk managers as the model for the markets. You have to think of the market as a leaky boat, and you have to keep working out where the 'holes' are and which will sink you, which are not an issue, and which can become bigger holes - even the unsinkable Titanic hit its iceberg.

So why not optimise and continually try and steer clear of the icebergs?

[BlueHorseshoe 10]

That's true for most developers because they fail to make the a priori choice to design an algorithm to guarantee specifically that it's "entirely eliminated".

 

Can you give an example of how you would suggest that this can be done? It's doesn't need to be anything useful or profitable - just an example of a 'non-optimisable system'.

 

BlueHorseshoe

[BlueHorseshoe 10]

You have to think of the market as a leaky boat, and you have to keep working out where the 'holes' are and which will sink you, which are not an issue, and which can become bigger holes - even the unsinkable Titanic hit its iceberg.

So why not optimise and continually try and steer clear of the icebergs?

 

I like this analogy a lot, not least because it suggests that sooner or later everybody's boat will sink.

 

So what is the 'optimal' time interval at which to re-optimise?

 

BlueHorseshoe

[SIUYA 11]

I like this analogy a lot, not least because it suggests that sooner or later everybody's boat will sink.

 

So what is the 'optimal' time interval at which to re-optimise?

 

BlueHorseshoe

 

I didn't mean that everyones boat will sink (efficient markets theories might suggest so) - you can have boats that go on and on and on so long as they are maintained , the point being --- to just plow full steam ahead might lead to you not recognising the dangers or thinking you have discovered the unsinkable truth - to always be keeping a watchful eye on things would be the prudent alternative (I have always been far to conservative)

 

Re when to re-optimise - with the advent of computers why not do it all the time - I would think that not much would change very much unless the market drastically changes.

I guess this might also depend on if the systems are mean reverting or trend following/momentum based. I kind of thought the machine learning ones basically do this anyway.

As mentioned these where more my random thoughts and I would re-optimise either pretty much continually if it was not a massive computer drain, or set a period to do it - ie;once a week, once a month, when volatility changed by a certain amount, or when there was some other trigger - eg; a move through a longer term MA maybe.

 

Was optimisation an issue or concern before computers, or is it just because we now have the tool we now have the problem? Did we just accept that a randomly generated MA did the trick before hand? Or did we kind of naturally optimise by using the eye?

[zdo 20]

Briefly -

Across time, the best times to re-optimize a system is at first degradation of performance.

 

Across time, the best optimizations are settings off those indicated by the re-optimization tests… mostly because the re-optimize is also now dated… how much off the optimization and which way to move the settings becomes a “projective art”

 

 

 

 

 

btw re: “I feel that knowing how to push winning trades is probably the best way to enhance a system”

Predictor you’re fixin’ to get one of my broken record lectures that’s usually reserved for beginners about discussing ‘best’ exit methods in the absence of identifying system type. If you are not system specific about your entries, don’t try to say what kind of stay is best…

 

In many systems, “push[ing] winning trades” is the ticket to suboptimal returns.

etc etc…

[BlueHorseshoe 10]

I didn't mean that everyones boat will sink (efficient markets theories might suggest so) - you can have boats that go on and on and on so long as they are maintained ,

 

I kind of thought the machine learning ones basically do this anyway.

 

Was optimisation an issue or concern before computers, or is it just because we now have the tool we now have the problem?

 

There's a good interview here with William Eckhardt in which he makes a good many relevant points: William Eckhardt: The man who launched 1,000 systems

 

One of the problems that I see with continual re-optimisation is that the optimal future parameters are always likely to be different from the current ones; therefore when a trader identifies an optimal current value for their system they are always better off chosing a parameter other than this to use.

 

One thing that I have been meaning to get around to researching for ages now is whether trends in optimal value over time can be identified and predicted in any way. As a very simple example, if the optimal value for an MA for trading the ES has fallen by ten over each of the past five months to a current optimal value of 50, is it possible to conclude that this trend will continue and thus predict the optimal value for the coming month as 40. This predicted optimum could then be used in live trading, rather than the current optimum of 50. One problem with this is local maxima within the space we explore when we backtest systems. Another problem is sample size - most 'AI' type strategies are thus forced to trade with relative high frequency so as to have available sufficient relevant information to respond to changes in the data in a way that is valid.

 

I think that before the widespread use of computers optimisation tended to be done from paper charts and calculations made by hand (must have taken forever) - for someone like Donchian there would have been no option. But then I have read accounts of people like Ed Sekoyta and Bill Dunn using computerised approaches for system design decades before anyone else . . .

 

I think that by far the best way to periodically re-optimise is 'by eye', as you remark, and this is effectively what a good discretionary trader will do. It requires experience and confidence though, and I for one have never been much good at it.

 

A few months ago I remarked on the trend in the ES at that time, saying that I considered it to be down (because the 'optimal' MA was at that time pointing downward). DB commented that the trend was clearly upward. He was right and I was wrong. I could see at the time that he was right - just looking at a chart it was clear that the trend was up. Having the confidence to act upon such convictions takes a certain type, however.

 

BlueHorseshoe

[BlueHorseshoe 10]

how much off the optimization and which way to move the settings becomes a “projective art”

 

We must have been typing at the same time, as I just repeated your point here below - sorry!

 

BlueHorseshoe

[SIUYA 11]

One of the problems that I see with continual re-optimisation is that the optimal future parameters are always likely to be different from the current ones; therefore when a trader identifies an optimal current value for their system they are always better off chosing a parameter other than this to use.

 

here is an idea - back test for say 1000 trades - get the optimal level - then backtest the last 400 trades - if the 1000 trade optimal level is say 50 MA, and the 400 trades is 60MA, then maybe the optimal level if you beleive in mean reversion will be to tend toward 59, then 58, then 57.....ie; always take two levels and use one slightly leaning toward the longer term avg.

I think that changing the risk management (qty) and stop or TP levels rather than just the MA might have more effect - eg; if 50MA is the longer term norm, and you are currently at 60, then it implies that the market either trends longer or is less volatile - so maybe changing the qty around based on deviations from some longer term MA might have more effect????

what about something like - when the MA blows out to 60, I will change the exit levels so they are different to when the market is what might be considered normal?

 

 

One thing that I have been meaning to get around to researching for ages now is whether trends in optimal value over time can be identified and predicted in any way. As a very simple example, if the optimal value for an MA for trading the ES has fallen by ten over each of the past five months to a current optimal value of 50, is it possible to conclude that this trend will continue and thus predict the optimal value for the coming month as 40. This predicted optimum could then be used in live trading, rather than the current optimum of 50. One problem with this is local maxima within the space we explore when we backtest systems. Another problem is sample size - most 'AI' type strategies are thus forced to trade with relative high frequency so as to have available sufficient relevant information to respond to changes in the data in a way that is valid.

 

This kind of flies in the face of mean reversion I would have thought.....and so if it did not fit with the general theory of how the market works, then you get caught in the trap of continually pushing something down until it rubberbands you.....maybe this is where some general stats can be useful - ie; normally it here, now we are lower than here, so odds are it will rebound, BUT in the meantime, the trend is your friend - run a tighter stop, or different TP levels.

Again, I think a lot goes back to the trade management, and the specific part of the market you are trying to capture.

If you are a day trader that goes both long and short intraday - then does it matter too much what the last months trend was - or are you only interested in the last couple of days for bias. I mean if you are really short term trading, does having 100 years of data help you?

 

It all a bit beyond me (the constant stats, models, testing - I understand enough to know that relying on them like they are the be all and end all is a fallacy - that is not to say you cant use them to make money) and I think chasing what is effectively a constantly changing type of market means that maybe the simplest thing is to have multiple systems that you either just rely on diversification to iron itself out, or discretion to avoid extremes....there is only one thing I know - if you buy it and it goes down in price, you loose money.....no matter what the model says.

[BlueHorseshoe 10]

It all a bit beyond me (the constant stats, models, testing - )

 

Don't worry - you're not missing out on anything if you can manage without them - they're dull as hell!

 

It's interesting that you mention relating entry/exit/market behaviour concepts to position sizing. This is something I've never been keen to do - the only element I tend to feed into a position sizing formula is the available equity. I think part of why I have preferred to keep the position sizing aspect seperate is that I know that the postion sizing formulas definitely "work", and for mathematical reasons they are certain to continue to do so - whereas my judgements about market behaviour may not be so accurate!

 

Something I've always wondered - did successful traders on the floor tend to use any concrete formula to size a position, or was it usually done on a discretionary basis?

 

BlueHorseshoe

[onesmith 10]

Whenever a trader designs a system/strategy etc, they make decisions which amount to a form of optimisation. The dangers of curve-fitting can be intelligently minimised; they cannot be eliminated entirely.

 

What do the words curve fitting mean?

 

The process of writing lots of strats without wasting time looking at their EC's will get you what you need. Symbolic logic skills to create anything imaginable. A catalog of strats in your head that answers all questions of whether bollinger beats something else without concern for their EC's because only their tangible value matters. Look for a place to make a break between associating edge with (immediately, instantly or easily attainable) profits. Make that break and you won't look back.

 

Maybe I'm mistaken but imo and with no disrespect intended if all that was behind you the peace of mind would enable you to rethink all the stuff that's taken for granted and write some simple tic tac toe type of algo comparable to the house edge in las vegas. When you get to the point where your questions are more philosophical like what is the objective of strategies in general and how can I acomplish that objective in the most straight forward way then you will create something outstanding worthy and unique. I know you have that in you and I know in time you will agree. If you're still interested in or thinking of the current question at that time you will be able to solve it multiple ways without semantics or trickery. Optimizable strats that eliminate rather than minimize curve fitting exist.

[SIUYA 11]

Don't worry - you're not missing out on anything if you can manage without them - they're dull as hell!

 

LOL - I did study econometirics at Uni as a requirement. Passed with a high mark as well, but hated them - it was at the same time when I had just started working on the floor, and so i would turn up to my finance lectures and they would tell me about these things in theory, and I would say "but in reality....." - one guy (thankfully) stopped me and said - do you want to pass the exam or not - I shut up, learnt it, passed it and then that was it......funny enough you then spend the rest of your life dealing with reality.

 

It's interesting that you mention relating entry/exit/market behaviour concepts to position sizing. This is something I've never been keen to do - the only element I tend to feed into a position sizing formula is the available equity. I think part of why I have preferred to keep the position sizing aspect seperate is that I know that the postion sizing formulas definitely "work", and for mathematical reasons they are certain to continue to do so - whereas my judgements about market behaviour may not be so accurate!

 

Something I've always wondered - did successful traders on the floor tend to use any concrete formula to size a position, or was it usually done on a discretionary basis?

 

BlueHorseshoe

 

Thats largely why a lot of these are just thoughts rather than tried and tested truisms that work until they dont - plus i am conservative enough to never really push things myself too much.

I would say that most used a discretionary formula that suited their risk profile, and adjusted it for what they percieved as liquidity and the prices at the time. Generally they were fairly consistent. ie; the bigger traders were always big, the smaller ones stayed small. This was also reliant on the types of stocks, and as most were trading options the option position.; eg; you could do a trade that cost $20k but might have made you $180k and the choice was to hedge or let it run. Market making was different to trading, and many traders were still just punters.

 

I would say the closest thing people got to a consistent formula was nothing like what a day trader used (% of equity etc per trade) - most either relied on their risk managers reigning them in (some people would bet the house every time if they could) , things like VAR, scenario risk limits, others (myself included) would determine roughly how big to let a book get based on how much risk you wanted in the greeks. We were also constrained by risk managers (our firm was very conservative and strict). Leverage was also another factor - you could run a big book with little costs and risk IF you balanced and hedged things carefully.

 

I guess we constrained ourselves for what we thought was the optimal size of the book that could be profitably controlled (without loosing you job, or too much sleep) over the long run.

 

Now when it came to running a position or having a naked outright position it might have been a different matter. Sometimes a position could be run that has very large if you had run it from a small cost. A lot may not have been too applicable here to day trading but the concept was the same......eg; if i could buy cheap options that i thought might make a lot of money i would do it all the time....most of the time it did not work out, but when it did - what fun. (I have made a lot of money out of options that where theoretically worth nothing, that then get sold for $1, $2,$3 - it was a handy thing it understand the power of letting a position compound itself when its going your way) Most day traders prefer to chip away, and there is nothing wrong with that, and the formulas work, but are maybe not as applicable here as this was options trading - concepts are the same, maths a bit different.

 

What was probably most interesting was the betting that was done for almost everything else - sports, backgammon, eating competitions. One guy i know became a profesional gambler - he was very strict when it came to that, but rather loose when it came to trading...go figure.

[zdo 20]

One guy i know became a profesional gambler - he was very strict when it came to that, but rather loose when it came to trading...go figure

 

"rather loose" works

'more loose' doesn't

[Momentum Chaser 10]

I like this thread.

 

I have found through much trial and error that the number or inputs (fixed optimizable variables) used in a strategy to create an entry trigger has a huge correlation to the sustainability of the system and the system's ability to perform well into the future. Fewer inputs = higher probabilty to continue to perform profitably is the premise that seems to work best for me.

 

Most of the strategies that I am currently writing and using use a maximum of 3 inputs, with only 2 inputs or 1 input being even better, in my opinion.

 

This tends to construct simpler strategies, that are more loose to the market, with the less possibility of being over curve fitted.

 

When looking to construct few input strategies, I tend to use inputs that might not be the first choices for many, since many basic indicators have more variables than I allow myself to use.

 

For example, most oscillators have a length, a high line and a low line and possibly other inputs that help to calculate the oscillating line. That totals 3 or more inputs, so I am maxed out. If a strategy had a moving average and an oscillator then I would be at 4 inputs minimum, which exceeds my guidelines.

 

Some inputs that are useful to me are Time of Day, Percent of a Range, On Balance Volume or other items that only have 1 input to calculate them. Percentage calculations tend to self adjust to changing market conditions better than fixed Length inputs.

 

Most of my systems are Always in, flip flop, breakout style systems. The idea is to always be in position to chase the new momentum of the market.

 

I agree with Suiya that the time to reoptimize is on a regular basis or when the market has changed and the system is begining to fail. The decision to be made at that time is whether the system is failing or the market is just not moving enough to be profitable with a breakout style system.

 

Momentum Chaser Steve

in San Diego

[christronic 10]

This is a great thread and something I have been interested in for a while now.

The is no sure way of determining how a system will perform in the long run

 

From what I have learnt the best way to test is walk forward optimization, and pick settings where the average setting around them produce a similar profit.

 

 

It seem most systems will perform badly at some point, the question is

 

How badly does a system have to perform before you stop using it.?

 

That's why I believe money management is very important and can make a poor

system perform ok.

 

Is money management Over Optimising?

I guess it could be, but it will stop you going broke.

[BlueHorseshoe 10]

A catalog of strats in your head that answers all questions of whether bollinger beats something else without concern for their EC's because only their tangible value matters.

 

I hope you don't think that I have any interest in Bollinger Bands just because I recently posted a few quick tests on here. They were for other people's possible benefit - I have no interest in things like Bollinger Bands at all . . .

 

BlueHorseshoe

[SIUYA 11]

Blue - have you ever tried testing something like this as an idea....

 

Take a MA, and for each MA level eg; 5, 10,15,20..... buy or sell a small amount.

combine them as if they were multiple different systems.

what are the results then?

 

You are then not really relying on a parameter - simply a series of them each with an equal weighting......

 

(just an idea, not sure how to apply it in EasyLanguage)

[onesmith 10]

I'm aware of sucessful traders that I have a deep respect for who claim bollinger bands are viable. My personal preferences are biased toward my own objective measures of the strengths and weaknesses of stuff so in that sense I prefer my subjective objective focus because it's that flavor of objectivity that works best for me. If you look hard enough at anything .. there's always a way someone hasn't thought of, so don't discount bollinger bands or anything else. Definitely don't pay any attention to anything except following your own strengths.

 

On the subject of optimization, it's worthwhile thinking of where curve fitting gets it's name. This is not an attempt to talk down and I realize others may be aware of this but sometimes the obvious is where the most value is and someone either might have missed it .. Or be able to correct what I say and thus contribute toward my understanding. The curve in curve fitting is Equity Curve. Any attempt to improve the curve itself will likely be fitting to something that's not viable. Over fitting is akin to overlaying 100 of these EC's on top of each other because visually we can then pick the best ec which is what the pc does during an optimization. There's no difference in curve fitting and over fitting. Neither of them are edge fitting. Edge fitting requires targeting an edge that is known to exist, while simultaneously outputing the success of that targeting in the form of signal strength. Overlaying those graphs is worthwhile eyeballing. It is also ok to iterate through them and let the pc select the most overfit edge fit because it's not looking at how it affects the ec, it's looking at it's success in finding what it's targeting.

[BlueHorseshoe 10]

From what I have learnt the best way to test is walk forward optimization, and pick settings where the average setting around them produce a similar profit.

 

I think that walk forward optimisation is useful, but not bulletproof. One of the difficulties with what you describe is that performance may or may not be normally distributed around the optimal value. So in the future, rather than the optimal value becoming one of those surrounding values that you describe, another local maxima swells to become the new optimum.

 

How badly does a system have to perform before you stop using it.?

 

That's a complicated question, and I don't have any kind of an answer to it. It will probably depend on your circumstances. A retail trader can open an account with a 10k balance and say 'I'm going to trade this through hell and high water, and I'll keep on trading until my broked tells me I don't have sufficient equity in the account to continue'. A fund, on the other hand, couldn't take that approach - there are too many restrictions on how well they must perform, maximum acceptable drawdowns, and the likelihood that investors will withdraw funds. Similarly, though you or I may be willing to do this with a 10k account, would you be comfortable doing it with a 200k account? I know that I wouldn't.

 

That's why I believe money management is very important and can make a poor

system perform ok.

 

Is money management Over Optimising?

I guess it could be, but it will stop you going broke.

 

A money management formula can be optimised/curve-fit to the historical risk associated with trading that strategy. Money management can only improve on what's there, so if by 'poor system performance' you mean a non-profitable performance, then it won't make its performance profitable. However, I too think that money-management is essential to long term success for most traders.

 

BlueHorseshoe

[zdo 20]

Just a couple of questions and comments about recent posts …and not disagreeing at all… these posts seem to be coming from knowledge, experience, and good (common) sense…

 

I think that walk forward optimization is useful, but not bulletproof. One of the difficulties with what you describe is that performance may or may not be normally distributed around the optimal value

Does it really matter that much if the distribution is normal or not? Isn’t the real issue that it is ‘distributed’ at all?

 

So in the future, rather than the optimal value becoming one of those surrounding values that you describe, another local maxima swells to become the new optimum.

ie Never use the local maxima. Move settings off it to the left or the right…and keep moving!

 

How badly does a system have to perform before you stop using it.?

Work on 'typical', conventional system design that uses specific and variable parameters has ‘smooth(er)’ Equity Curve as a main requirement… so if ‘typical’ is the operative term for you, first rule would be to stop using a system if the EC is too ‘jagged and ragged’.

Regarding the direction of EC, one way to handle it is to predetermine a reasonable, (and slightly generous) X number of ‘reporting periods’ for the system the EC will be allowed to go downward before stopping the system (and also to predetermine a number X + i of ‘reporting periods’ the curve will be allowed to go flat before stopping). This way is more general and has wider applicability than does using generalized % drawdown, etc criteria – ie any % drawdown criteria used should be VERY system specific

 

money management …it will stop you going broke.

christronic if I could make a suggestion, you might serve your own purposes best by going back and reforming, restructuring, redefining etc you’re whole concept of money management

The MM process must be clearly demarcated and always after completion of any ‘expectation’ and ‘optimization’ work on a system and then it can be narrowed down to one thing – Position sizing!

Every system with a 'positive expectation' has a sweet spot to the left of optimum-f (using stop size, not worst loss, in calcs btw). A good seed starting place for ‘typcial’ systems is around 20% left of Opt-f peak…. then dynamically tweak indiv. system from there

 

This whole ‘systems optimization’ business is an art really… with many books and articles (and sites and posts) explaining how it’s pure science of course :rofl:

… reminds me of what SUIYA was saying above…

turn up to my finance lectures and they would tell me about these things in theory, and I would say "but in reality....." - one guy (thankfully) stopped me and said - do you want to pass the exam or not - I shut up, learnt it, passed it and then that was it......funny enough you then spend the rest of your life dealing with reality.

concurring - In my reality, I go through phases where I run and look at a many bunches of stats … and about 1 in 1000 reveal something surprising / differentiating / useful…

 

 

 

 

 

 

 

 

Fwiw, and hopefully on topic, I would never run just one of these 'optimizable' ( :haha: like there are types that aren't 'optimizable') types of systems… always a ‘portvoleo’ of at least four systems of varying complexity (see Momentum Chaser’s post # 19 above)…and with one of them ‘completely’ negatively correlated... just saying... experience from days of fund management / smooth, correctly sloping EQ's ...

 

This was a great week to be trading!

Have a great weekend all.